# Welcome
### You’re in the official guide for **Chat Video Pro**
#### **What you’ll find here:**
• **Getting Started** — Install, activate, and start using ChatVideoPro with confidence.\
• **Feature Guides** — Learn how each assistant and tool works, with examples you can try right now.\
• **Actionable Workflows** — Real editing scenarios that show how to save hours on common tasks.\
• **Troubleshooting & FAQ** — Quick answers to common questions so you stay in the flow.
We recommend starting with **the guides below.** Happy editing! 🎥
### Quick Links
### Need Help?
* Check the Troubleshooting section
* Review Common Workflows
* Explore Model Documentation for specific features
**Want your workflow built, not just explained?** Most editors spend 2–3 weeks figuring out what we cover in 60 minutes. Book a Founder Onboarding Session — a live build session where we configure Chat Video Pro around your real project. Come with Premiere open. Leave with a working AI pipeline inside your timeline. [BOOK YOUR SESSION →](https://checkout.chatvideopro.com/checkout/buy/6b983504-8882-45e4-9ca8-64ea1a28eddf)
Ready to start? Pick up your copy today at [www.chatvideopro.com](https://www.chatvideopro.com)
# What is Chat Video Pro?
Chat Video Pro is an AI-powered video editing assistant that runs directly inside Adobe Premiere Pro. It combines the power of generative AI with professional video editing workflows, allowing you to
{% embed url="" %}
{% embed url="" %}
## Core Concept
Think of Chat Video Pro as your intelligent co-editor that lives inside Premiere Pro. Instead of switching between multiple applications or websites, you can:
* **Generate videos** from text descriptions using cutting-edge AI models (Sora 2, Veo 3.1, Kling, Hailuo 03, Grok Imagine 1.5, WAN)
* **Create images** for thumbnails, graphics, or video elements
* **Get color grading advice** with technical analysis and downloadable LUT files
* **Cut stories** from long interviews using AI-powered transcript analysis
* **Remove backgrounds** from images and videos with professional precision
* **Apply visual effects** to existing footage
* **Get Premiere Pro help** with troubleshooting and workflow questions
Learn more at [www.chatvideopro.com](https://www.chatvideopro.com)
### Key Features
#### Video Generation
{% content-ref url="/pages/ZabdRu4lR5kM24lY1jbP" %}
[Video Generation](/features/video-generation)
{% endcontent-ref %}
#### Image Generation & Editing
{% content-ref url="/pages/BVewVtsEfeYnOAEvB0JY" %}
[Image Generation](/features/image-generation)
{% endcontent-ref %}
#### Pre-Trained Conversation Starters
{% content-ref url="/pages/vEilOoYPBivQ6ii2S9Wx" %}
[Story Cutter Assistant](/conversation-starters/story-cutter-assistant)
{% endcontent-ref %}
{% content-ref url="/pages/PtL6vG4vyqifUC1SfNHl" %}
[Color Grade Assistant](/conversation-starters/color-grade-assistant)
{% endcontent-ref %}
{% content-ref url="/pages/b2RhKLpaZ2euWElGlMRV" %}
[Brand Voice Assistant](/conversation-starters/brand-voice-assistant)
{% endcontent-ref %}
{% content-ref url="/pages/ZfUvhZA5MYV4yglgwipM" %}
[Video Prompter Assistant](/conversation-starters/video-prompter-assistant)
{% endcontent-ref %}
### How It Works
Chat Video Pro runs as a CEP (Common Extensible Platform) extension inside Premiere Pro. It uses:
* **Fal.ai API** for all generative media operations
* **Local processing** for frame analysis, transcript chunking, and memory storage
All your API key and data stay on your machine—nothing is sent to third-party servers except for the actual generation requests.
### Who Is It For?
Chat Video Pro is designed for:
* **Video editors** who want to speed up their workflow
* **Content creators** producing YouTube, TikTok, Instagram, and other social media content
* **Professional editors** working on documentaries, commercials, or narrative projects
* **Anyone** who wants to leverage AI for video creation without leaving Premiere Pro
### What Makes It Different?
Unlike standalone AI video tools, Chat Video Pro:
* **Lives inside Premiere Pro** - No context switching
* **Pay-as-you-go model** - You only pay for what you create
* **Integrates with your timeline** - Export frames, import clips directly
* **Professional workflows** - LUT files, ProRes output, Lumetri integration
* **Multiple AI models** - Choose the best model for each task
* **Conversational interface** - Natural language commands, not just buttons
***
**New to Chat Video Pro?** Don't spend days figuring out what takes 60 minutes with the founder. Book a Founder Onboarding Session: we open Premiere together, build your AI workflow around how you actually edit, and you leave running. [BOOK YOUR SESSION →](https://checkout.chatvideopro.com/checkout/buy/6b983504-8882-45e4-9ca8-64ea1a28eddf)
**Next Steps:** Check out Installation & Activation to get started, or explore First 5 Things to Try for a quick start guide.
# Pricing
Chat Video Pro uses a license-based model with pay-per-use generation costs through Fal.ai.
{% embed url="" %}
**Want it done for you?** Skip the learning curve — book a Founder Onboarding Session and we'll build your AI workflow live inside Premiere. 60 minutes, your project, done. [BOOK YOUR SESSION →](https://checkout.chatvideopro.com/checkout/buy/6b983504-8882-45e4-9ca8-64ea1a28eddf)
## License Cost
Chat Video Pro requires a one-time license purchase. Multiple seats can be purchased if you'd like to use the software on more than one device.
#### Where to Purchase
* Purchase directly from **[www.chatvideopro.com](http://www.chatvideopro.com)**
* License keys are delivered via email after purchase
### Generation Costs (Fal.ai Credits)
All video, image, and AI operations run through **Fal.ai**, which uses a pay-as-you-go model. Meaning you only pay for what you create. To set up, you'll need to:
1. **Create a Fal.ai account** at [fal.ai](https://fal.ai)
2. **Add credits** to your Fal.ai account
3. **Provide your Fal API key** (`key_id:key_secret`) during Chat Video Pro setup
4. You will see all usage from inside the usage dashboard in Chat Video Pro
{% content-ref url="/pages/8teMzAnXOweDGrYD9JBK" %}
[Usage Panel](/getting-started/interface-overview/usage-panel)
{% endcontent-ref %}
#### Cost Estimates
Costs vary by model and operation type and can change over time:
| Operation | Model | Approximate Cost |
| ---------------------- | --------------------------------------- | -------------------------------------------------------------- |
| Story Cutter | Claude Sonnet 5 |
|
| **Text-to-Video** | Seedance 2 | \~$0.30 per second at 720p (higher at 1080p / 4K) |
| **Text-to-Video** | Seedance 2 Mini | \~$0.07 per second at 480p; \~$0.15 per second at 720p |
| **Text-to-Video** | Google Omni Flash | \~$0.125 per second at 720p |
| **Text-to-Video** | Grok Imagine Video | \~$0.05 per second at 480p; \~$0.07 per second at 720p |
| **Text-to-Video** | Grok Imagine Video 1.5 (Image-to-Video) | \~$0.08 per second at 480p; \~$0.14 per second at 720p |
| **Text-to-Video** | Veo 3.1 | \~$0.04-0.12 per second |
| **Text-to-Video** | Kling 3.0 | \~$0.03-0.10 per second |
| **Text-to-Image** | Flux 2 Max | \~$0.01-0.03 per image |
| **Text-to-Image** | Seedream 5.0 Pro | \~$0.07 per image |
| **Text-to-Image** | Ideogram V4 Fast | \~$0.01 per image |
| **Image Upscaling** | Topaz | \~$0.01-0.05 per image |
| **Video Upscaling** | Topaz | \~$0.10-0.50 per video |
| **Background Removal** | Bria RMBG | \~$0.001-0.01 per image |
| **SAM 3 Rotoscoping** | SAM 3 | \~$0.05-0.20 per second of video |
What This Means in Real Usage For a typical **2-hour transcript** (StoryCutter workflow):
* Input size: \~25K tokens
* Output (edited story cut): \~5K-12K tokens
Estimated Cost per Edit:
* Claude Sonnet 5: \~$0.15 - $0.30
* GPT-5.5: \~$0.12 -$0.25
* Gemini 3.1 Pro: \~$0.10 -$0.22
Most story cuts cost under $0.30 total
Key Takeaway\
Even long-form edits (2+ hours of footage) cost just\
a few cents to process, making Al-powered story\
cutting extremely scalable for high-volume\
workflows.
{% hint style="info" %}
**Note:** These are approximate costs from fal.ai model pages. Seedance and Omni Flash bill by tokens; per-second figures are fal's published 720p estimates. Check [fal.ai/pricing](https://fal.ai/pricing) for current rates.
{% endhint %}
#### Managing Costs
* **Monitor usage** in Chat Video Pro: Gear icon → Usage panel
* **View detailed billing** at [fal.ai/dashboard/usage-billing](https://fal.ai/dashboard/usage-billing)
* **Set budget alerts** in your Fal.ai account settings
* **Use preview modes** when available (e.g., SAM 3 "Track Frame" before processing full video)
### Refund Policy
We offer a seven-day money-back guarantee for the base package of Chat Video Pro:
* **In-app:** Gear icon → Contact Us
* **Email:** Support contact information provided in the About panel
### Pricing FAQ's
For any questions not covered here please review the dedicated pricing FAQ page.
{% content-ref url="/pages/mO6B86vbSZ8TMhl9eO7v" %}
[Pricing FAQ](/troubleshooting-and-faq/pricing-faq)
{% endcontent-ref %}
***
**Next:** Learn about Compatibility requirements before installing.
# Compatibility
Before installing Chat Video Pro, ensure your system meets the following requirements.
## Adobe Premiere Pro
#### Supported Versions
Chat Video Pro requires **Adobe Premiere Pro 2024 or later** (version 24.0+).
| Premiere Pro Version | Status | Notes |
| -------------------- | ----------------- | ----------------------------------- |
| **2026** (v26.x) | ✅ Fully Supported | Latest — Recommended |
| **2025** (v25.x) | ✅ Fully Supported | Minimum supported version |
| **2024 and earlier** | ❌ Not Supported | CEP extension requirements not met. |
#### How to Check Your Version
1. Open Premiere Pro
2. Go to **Help → About Adobe Premiere Pro**
3. The version number appears in the dialog
### Operating System
#### Windows
* **Windows 10** (64-bit) or later
* **Windows 11** (64-bit) - Recommended
#### macOS
* **macOS 10.15** (Catalina) or later
* **macOS 11+** (Big Sur, Monterey, Ventura, Sonoma) - Recommended
### System Requirements
#### Minimum Requirements
* **RAM:** 8 GB (16 GB recommended for video generation)
* **Storage:** 500 MB free space for extension installation
* **Internet:** Stable connection required for AI generation
#### Recommended Requirements
* **RAM:** 16 GB or more
* **Storage:** 2 GB+ free space (for generated media cache)
* **Internet:** High-speed connection (for faster generation downloads)
### Additional Software
#### Required
* **Adobe Media Encoder** (comes with Premiere Pro)
* Required for exporting nested sequences and adjustment layers via Clip Export
* Automatically used when needed
### API Requirements
#### Fal.ai Account
* **Required:** Active Fal.ai account with credits
* **API Key Format:** `key_id:key_secret`
* **Where to Get:** [fal.ai/dashboard](https://fal.ai/dashboard)
### Network Requirements
#### Required
* **Internet connection** for all AI generation operations
* **HTTPS access** to fal.ai API endpoints
#### Firewall Considerations
If you're behind a corporate firewall, ensure these domains are accessible:
* `fal.ai` and all subdomains
### Checking Compatibility
Before installing:
1. ✅ Verify Premiere Pro version (2024+)
2. ✅ Check operating system compatibility
3. ✅ Ensure stable internet connection
4. ✅ Create Fal.ai account and add credits
5. ✅ Have your license key ready
***
**Next:** Proceed to Installation & Activation to set up Chat Video Pro.
# Installation & Activation
Follow these steps to install and activate Chat Video Pro in Adobe Premiere Pro. The entire process takes just a few minutes.
## Setup Video
{% embed url="" %}
## Step 1: Download Chat Video Pro
#### Check Your Email
After purchase, you'll receive an email with your Chat Video Pro receipt containing:
* ✅ **Download link** for the Chat Video Pro `.zxp` plugin file
* ✅ **Your personal license key** to unlock Chat Video Pro
**Can't find the email?**
* Check your spam or promotions folder
* Look for an email from Momentohm Media or your payment processor
* Contact support if you still can't locate it
#### Access Your Download
1. **Open your receipt email**
2. **Scroll down** to find the "View Order" button
3. **Click "View Order"** to access your order page
4. **Download both:**
* **ChatVideoPro.zxp** - The extension file
* **AEScripts ZXP Manager** - The installer tool (Mac or Windows) [AEScripts ZXP Installer](https://aescripts.com/learn/post/zxp-installer/)
### Step 2: Install the Plugin
#### Using aescripts ZXP Installer
1. **Open the AEScripts ZXP Installer**
* Launch the ZXP Manager application you downloaded
2. **Drag and Drop**
* Drag the **ChatVideoPro.zxp** file into the ZXP Manager window
* You should see it appear in the installer interface
3. **Click Install**
* Click the **Install** button in the ZXP Manager
* Wait for the installation to complete
* You'll see a confirmation when Chat Video Pro is installed
#### Open in Premiere Pro
1. **Launch Adobe Premiere Pro**
2. **Open the Extension**
* Go to **Window → Extensions → Chat Video Pro**
* The panel will open (you can dock it wherever you prefer)
3. **Onboarding starts automatically**
* The activation wizard will appear
### Step 3: Activate Your License
#### A) Activate Your License Key
1. **Copy your license key**
* Find it in your receipt email
2. **Paste into Chat Video Pro**
3. **Click "Next"**
#### B) Create Your FAL Key
Chat Video Pro uses your own FAL.ai API key
1. **Click "Get FAL Key"**
* It opens [fal.ai/dashboard/keys](https://fal.ai/dashboard/keys) in your browser
2. **Sign in to FAL.ai**
* Sign in with your Google account (or create a new account)
3. **Create a New Key**
* Once logged in, click **"Add Key"** or **"Create Key"**
* Choose **"Admin Key"** type
* Give it a name like "Chat Video Pro" or "CVP"
* Click **"Create Key"**
4. **Copy Your Key**
* The key will appear in the format: `key_id:key_secret`
* **Important:** Copy the entire key (both parts with the colon)
* Save it somewhere safe so you don't have to repeat this process
5. **Paste into Chat Video Pro**
* Return to the Chat Video Pro activation screen
* Paste your FAL key into the field
* Click "Next" or "Save"
#### C) Add Funds to Your FAL Account
Before you can generate videos, you need to add funds to your FAL.ai account.
1. **Access Usage Dashboard**
* In Chat Video Pro, click the **Gear icon** (⚙️) in the header
* Select **Usage** from the menu
2. **View Your Balance**
* Click **"See Balance"** or **"View in Fal Dashboard"**
* This opens your FAL.ai account dashboard
3. **Add Funds**
* Recommended starting amount: **$10-$20**
* Add more if you plan to generate many videos
* FAL uses a pay-as-you-go model with no subscription
**Why add funds?**
* Chat Video Pro uses a **pay-as-you-go model**
* No subscription fees
* No upcharges
* No credits that expire
* Track spending per project
* Only pay for what you use
### You're Ready to Create!
Once activation is complete, you now have access to:
* ✅ **AI Video Models** (Sora, Veo, Kling, Hailuo 03, Grok Imagine 1.5, WAN)
* ✅ **Upscaler Tools** (Topaz, Flash VSR 2, Bria)
* ✅ **Canvas Editor** (Image editing and manipulation)
* ✅ **ChatGPT Assistant** (Workflow help and troubleshooting)
* ✅ **Story Cutter** (Transcript analysis and paper cuts)
* ✅ **All inside your Premiere timeline**
#### Next Steps
* **Dock the panel** next to your Lumetri scopes for easy access
* **Watch the full walkthrough:** [YouTube Tutorial](https://youtu.be/3tlWda8jsf4)
* **Join the Discord:** [Chat Video Pro Community](https://discord.gg/WjGaUYtNyr)
* **Try your first generation:** See First 5 Things to Try
### Troubleshooting
#### "I don't see Chat Video Pro inside Premiere"
**Solution:**
1. **Verify installation** - Check that the ZXP file was successfully installed
2. **Check Premiere Pro version** - Must be 2026 or later
3. **Restart Premiere Pro** - Close and reopen the application
4. **Check Extensions menu** - Go to Window → Extensions → Chat Video Pro
5. **Enable CEP debugging** (if still not visible):
**Windows:**
* Create registry key: `HKEY_CURRENT_USER\Software\Adobe\CSXS.11` (or CSXS.10, CSXS.9 for older Premiere)
* Add DWORD: `PlayerDebugMode` = `1`
* Restart Premiere Pro
**macOS:**
```bash
defaults write com.adobe.CSXS.11 PlayerDebugMode 1
```
(Change `11` to your Premiere version number)
* Restart Premiere Pro
#### "My license key won't activate"
**Solutions:**
* Verify you copied the **entire key** (no extra spaces before/after)
* Check the key is from a **valid purchase**
* Ensure you're using the key from your **receipt email**
* Try copying the key again (sometimes hidden characters get included)
* Contact support at **** if issues persist
#### "Where do I paste my API key?"
**FAL Key:**
* Paste it in the activation wizard during setup
* Or later: Gear icon → Settings → API Keys → FAL.ai Key field
* Format: `key_id:key_secret` (both parts with colon)
**Other API Keys (Optional):**
* OpenAI: Gear icon → Settings → API Keys → OpenAI
* Gemini: Gear icon → Settings → API Keys → Gemini
#### "How do I add funds for video generation?"
1. Click **Gear icon** (⚙️) → **Usage**
2. Click **"See Balance"** or **"View in Fal Dashboard"**
3. This opens your FAL.ai account
4. Click **"Add Funds"** or **"Add Credits"**
5. Recommended: Start with $10-$20
6. Funds never expire - pay only for what you use
#### "Do I need a subscription?"
**No!** Chat Video Pro uses a **pay-as-you-go model**:
* ✅ No subscription fees
* ✅ No monthly charges
* ✅ No credits that expire
* ✅ Only pay for generations you use
* ✅ Track spending per project
#### FAL Key Errors
**Common issues:**
* **Format incorrect:** Must be `key_id:key_secret` (with colon, both parts)
* **Key inactive:** Verify the key is active in your FAL dashboard
* **No funds:** Add credits to your FAL account (see Usage panel)
* **Wrong key type:** Make sure you created an "Admin Key" in FAL
**To fix:**
1. Go to [fal.ai/dashboard/keys](https://fal.ai/dashboard/keys)
2. Verify your key is active
3. Copy the full key again (both parts)
4. Paste into Chat Video Pro: Gear → Settings → API Keys
### Privacy & Security
* **All keys stored locally** - Never sent to external servers except for API calls
* **License validation is local** - No phone-home for activation
* **Generated media cached locally** - Stored in your chat media folders
* **Transcripts and memory** - All stored in local browser storage
* **Your data stays private** - We never touch your data or API keys
### Getting Support
* **Email:**
* **Discord:** [Join the Community](https://discord.gg/WjGaUYtNyr)
* **In-app:** Gear icon → Contact Us
* **Founder Onboarding Session:** Not a support call — a live 60-minute session where we build your AI workflow inside your actual Premiere project. Come with a project open; leave with a working pipeline. [BOOK YOUR SESSION →](https://checkout.chatvideopro.com/checkout/buy/6b983504-8882-45e4-9ca8-64ea1a28eddf)
***
**Next:** Learn about the Interface Overview to understand Chat Video Pro's layout.
# How to Update Chat Video Pro
This guide walks you through updating Chat Video Pro to the latest version inside Adobe Premiere Pro.
Updating ensures you have:
* The latest features
* Bug fixes and performance improvements
* Compatibility with newer models and workflows
> **Important:** Chat Video Pro updates require uninstalling the previous version before installing the new one. This is normal for CEP extensions and does not affect your license.
### Before You Begin
Make sure:
* Adobe Premiere Pro is **closed**
* You have access to the **ZXP installer** you originally used
* You are logged into Chat Video Pro to download the latest version
Your license and settings are not affected by updating.
***
### Step 1: Uninstall the Current Version
1. Open the **ZXP Installer** you used when first installing Chat Video Pro
2. Click **Premiere Pro**
3. Select **Chat Video Pro**
4. Click **Uninstall**
Wait until the uninstall process completes before moving on.
> If Premiere Pro is open during uninstall, the process may fail.
***
### Step 2: Install the Latest Version
1. Open **Chat Video Pro** (before uninstalling, if needed)
2. Go to the **Gear icon → Updates → Download**
3. Download the latest `ChatVideoPro.zxp` file
4. Open the **ZXP Installer**
5. Drag and drop the new `.zxp` file into the installer
6. Confirm the installation
7. Done ✅
Once installed, you can launch Adobe Premiere Pro and open Chat Video Pro as usual.
***
### After Updating
After restarting Premiere Pro:
* Open Chat Video Pro
* Confirm the version number **Gear icon → Updates**
* Test a simple generation to confirm everything is working
If you experience issues, see:
* **Installation & Setup FAQ**
* **Generation Errors & Failed Jobs FAQ**
***
### Common Update Questions
#### **Do I need to uninstall before updating?**
Yes. Chat Video Pro updates require uninstalling the previous version first. Installing over an existing version may cause issues.
***
#### **Will updating remove my license or settings?**
No. Your license remains valid. In rare cases, you may be prompted to re-enter your API key.
***
#### **What if the new version doesn’t appear in Premiere Pro?**
Try:
1. Restarting Premiere Pro
2. Restarting your computer
3. Reinstalling the `.zxp` file
If the issue persists, see the **Installation FAQ** or contact support.
***
#### **How often should I update?**
We recommend updating whenever a new version is released to ensure:
* Best performance
* Latest features
* Continued compatibility
***
### Need Help Updating?
If you run into issues:
* Open Chat Video Pro → **Gear icon → Contact Us**
* Or email support
Include:
* Your OS (macOS or Windows)
* Premiere Pro version
* What step you’re stuck on
***
**Next:** Learn how to track changes and usage in the [**Usage Dashboard**](/getting-started/interface-overview/usage-panel).
# First 5 Things to Try
New to Chat Video Pro? Start here with these essential workflows that showcase the tool's core capabilities.
### Quick Wins
#### Chat Video Pro can do a lot, but you do not need to learn everything at once.
Start with these five workflows. Together, they show the main shape of the tool:
* Studio for guided creative workflows.
* Generate Media for direct model control.
* Conversation Starters for specialized assistants.
* Timeline tools for frame capture and clip import.
* Library/results reuse so one generation can feed the next.
{% embed url="" %}
***
#### 1. Create a Cinematic Still in Studio
**Goal:** Turn an idea into a polished image you can use as a thumbnail base, key art, concept frame, reference image, or source frame for video.
Use this first because it teaches the most important Studio idea: design the frame before you animate or edit it.
**Steps**
1. **Open Studio**
* Click **Studio** in the sidebar.
* Choose [**Cinematic Lab**.](/features/studio/cinematic-lab)
2. **Describe the scene**
* Include subject, setting, mood, and visual style.
* Example: `A cinematic close-up of a chef plating a dessert in a moody restaurant kitchen, warm practical lights, shallow depth of field.`
3. **Choose the creative controls**
* Pick aspect ratio based on the final use.
* Choose a camera/lens direction if available.
* Try a high-quality image model when the still needs to carry the whole idea.
4. **Generate a batch**
* Review the grid.
* Pick the strongest image.
* Send it back to chat or reuse it in another Studio workflow.
**Pro tip:** Treat Cinematic Lab like a mini lookdev session. If the final video should feel premium, make the still premium first.
See Studio Cinematic Lab.
***
#### 2. Animate a Still with Motion Director
**Goal:** Turn one strong image into a controlled moving shot.
This is the fastest way to understand how Studio workflows connect. A Cinematic Lab result can become the source image for Motion Director.
**Steps**
1. **Open Studio**
* Choose [**Motion Director**](/features/studio/motion-director).
2. **Load a source image**
* Use an image from Recents, the Library, file upload, or a frame you captured from Premiere.
3. **Choose a camera movement**
* Try a push in, dolly, orbit, crane, handheld, drone, or reveal move.
4. **Add direction**
* Keep the prompt focused on motion and continuity.
* Example: `Slow dolly push toward the subject, subtle parallax, preserve the character identity and lighting.`
5. **Generate**
* Review the moving shot.
* Use the result in your edit or try another motion preset.
**Pro tip:** Start with the final framing you want. Motion Director can animate a still, but it cannot save a source image that is cropped too tightly.
See Studio Motion Director.
***
#### 3. Generate a Video Directly with Generate Media
**Goal:** Create a new video from a written prompt while controlling the model and settings yourself.
Use Generate Media when you want direct control over the model, aspect ratio, duration, resolution, references, and audio settings.
**Steps**
1. **Enable Generate Media**
* Click the **Generate Media** toggle in the composer.
* The model selector and settings appear.
2. **Select a video model**
* Choose the model based on the shot.
* Use the model guide if you are not sure.
3. **Set your parameters**
* **Aspect ratio:** Match the destination, such as 16:9 for YouTube or 9:16 for vertical.
* **Duration:** Start short while testing.
* **Audio:** Enable only when the selected model supports it and the shot benefits from sound.
4. **Write the prompt**
* Include subject, setting, action, camera, lighting, and style.
* Example: `Wide cinematic shot of a mountain road at sunrise, slow drone push forward, warm golden light, mist in the valley, realistic documentary style.`
5. **Generate**
* Review the result.
* Regenerate, edit, extend, or reuse the clip as needed.
**Pro tip:** If the shot needs to match your edit, capture a frame from your timeline and use it as visual context before generating.
See Text-to-Video and Supported Video Models.
***
#### 4. Cut a Story from a Transcript
**Goal:** Turn a long interview, podcast, webinar, or documentary transcript into an edit-ready paper cut.
Story Cutter is one of the best first workflows because it shows how Chat Video Pro can help with editing decisions, not just media generation.
**Steps**
1. **Open Story Cutter Assistant**
* Click the **Story Cutter Assistant** conversation starter.
2. **Export your transcript from Premiere Pro**
* Open the **Text** panel in Premiere Pro.
* Transcribe your footage if needed.
* Use the panel menu to export the transcript as Premiere Pro's native `.json` transcript file.
3. **Attach the transcript**
* Use the paperclip button in Chat Video Pro.
* Make sure the `.json` transcript is attached before asking for a cut.
4. **Describe the edit**
* Include platform, target runtime, tone, audience, and story goal.
* Example: `Create a 60-second YouTube Shorts cut with a strong hook, fast pacing, and one clear takeaway.`
5. **Review and refine**
* Ask for shorter versions, stronger hooks, alternate structures, or a batch of cuts.
**Pro tip:** SRT, VTT, and plain text transcripts are not the recommended path. Use Premiere Pro's `.json` transcript export for best results.
{% embed url="" %}
***
#### 5. Import a Timeline Clip and Improve It in Studio
**Goal:** Take footage from your Premiere timeline and use AI to clean, transform, or finish it.
This shows the Premiere-native side of Chat Video Pro: your existing edit can feed Studio without leaving the project.
**Steps**
1. **Select a clip in Premiere Pro**
* Trim to the section you actually want to process.
* If you need Premiere effects or color included, nest the clip first.
2. **Click Import Clip**
* The clip is exported into Chat Video Pro as a video attachment.
3. **Open Studio**
* Choose the workflow that matches the job:
| Goal | Studio workflow |
| ------------------------------------- | --------------- |
| Remove a subject background | Rotoscope |
| Remove an object or distraction | Erase Objects |
| Add rain, fire, fog, energy, or style | Add Effects |
| Change part of a shot | Reshoot |
| Improve final resolution | Upscale |
| Change lighting or mood | Relight Scene |
4. **Run a short test**
* Use the shortest useful segment first.
* Review before processing more.
5. **Send the result back to the edit**
* Save, download, or reuse the result from the Library.
**Pro tip:** Upscale last. Run creative changes first, approve the shot, then upscale the final version.
See Import Clip Button and Studio.
***
#### Bonus: Get Color or Premiere Help
If you want a quick non-generation win, try one of these:
* Use Color Grade Assistant with a captured frame and ask, `What is wrong with this shot?`
* Ask Premiere Pro Guru, `How do I nest this sequence?`
* Use Thumbnail Mode to generate a thumbnail direction from your edit.
* Use Background Removal to make a transparent cutout from a still image.
***
#### What's Next?
After these five workflows, you will understand the main map:
* [**Studio**](/features/studio) is for guided creative and post-production workflows.
* [**Generate Media**](/getting-started/interface-overview/generate-media-button) is for direct image and video model control.
* [**Conversation Starters**](/getting-started/interface-overview/conversation-starters) are specialized assistants.
* [**Frame Capture**](/getting-started/interface-overview/frame-capture-button) and [**Import Clip**](/getting-started/interface-overview/import-clip-button) connect your Premiere timeline to AI tools.
* [**Library** ](/getting-started/interface-overview/library)lets you reuse results across workflows.
Next pages to read:
* Studio
* Generate Media Button
* Model Selection
* Generation Errors & Failed Jobs FAQ
**Skip the trial and error.** Most editors spend 2-3 weeks figuring out what we cover in 60 minutes. Book a Founder Onboarding Session and we will build your AI workflow live inside your Premiere timeline. [BOOK YOUR SESSION ->](https://checkout.chatvideopro.com/checkout/buy/6b983504-8882-45e4-9ca8-64ea1a28eddf)
# Interface Overview
Chat Video Pro's interface is designed to feel familiar while providing powerful AI capabilities. Here's a complete guide to every element.
### **Walkthrough Video**
{% embed url="" %}
## Main Panel Layout
The Chat Video Pro panel consists of several key areas:
### Chat Header
Located at the top of the panel, the header contains:
#### Gear Icon (⚙️)
Click to access:
* **Settings** - API keys, microphone, conversation starters, AI learning
* **Usage** - Fal.ai usage statistics and billing links
* **Updates** - Check for new versions and view changelog
* **Debug Console** - Real-time logs (for troubleshooting)
* **About** - Version info and reset extension option
* **Contact Us** - Support channel chooser
### Conversation Area
The main message display area where:
* **Your messages** appear on the right (blue)
* **AI responses** appear on the left (gray)
* **Generated media** (images/videos) display inline
* **Thinking cards** show multi-step process progress
* **Tool results** appear as cards (LUT files, analysis, etc.)
#### Message Actions
Hover over any message to see:
* **Regenerate** - Re-run the AI response
* **Copy** - Copy message text
* **Edit** - Edit and resend your message
* **Delete** - Remove the message
### Composer (Input Area)
The bottom section where you type, upload files, and control generation.
#### Text Input
* **Main textarea** - Type your messages here
* **Auto-expands** as you type longer messages
* **Supports markdown** in some contexts
#### Toolbar Buttons
From left to right:
**1.** [**Frame Export Button**](/getting-started/interface-overview/frame-capture-button) **(📷)**
* **What it does:** Captures the current frame from your Premiere Pro timeline
* **When to use:** Color grading, analysis, or using timeline footage as reference
* **Output:** Image appears in composer as attachment
**2.** [**Clip Export Button**](/getting-started/interface-overview/import-clip-button) **(🎥)**
* **What it does:** Exports the selected clip from your timeline
* **When to use:** Video-to-video editing, SAM 3 rotoscoping, or sending footage to AI
* **Output:** Video thumbnail appears in composer
* **Limits:** Max 1 GB, 5 minutes duration
**3.** [**Voice Input Button**](/getting-started/interface-overview/voice-input) **(🎤)**
* **What it does:** Records voice input and transcribes to text
* **When to use:** Quick prompts, hands-free operation
* **Setup:** Select microphone in Settings → Microphone
**4. File Upload Button (📎)**
* **What it does:** Opens file picker for images, videos, transcripts, PDFs
* **Supported:** Images (JPG, PNG, WebP), Videos (MP4, MOV), Text files (.txt), PDFs
* **Drag & drop** also works directly onto the composer
**5.** [**Generate Media Toggle**](/getting-started/interface-overview/generate-media-button)
* **What it does:** Switches between chat mode and media generation mode
* **When enabled:** Shows model selector, aspect ratio, duration controls
* **When disabled:** Standard chat interface
#### Generate Media Mode
When the toggle is enabled, additional controls appear:
**Model Selector**
* **Video Models:** Sora 2, Veo 3.1, Seedance 2, Kling 3.0, Hailuo 03, Grok Imagine 1.5, Wan 2.7, etc.
* **Image Models:** Flux 2 Max, GPT Image 2, Nano Banana Pro, etc.
* **Video-to-Video:** SAM 3, Kling VFX, LTX Reshoot (when video attached)
**Media Settings Pills**
* **Aspect Ratio:** 16:9, 9:16, 1:1, 4:3, etc.
* **Resolution:** 720p, 1080p, 4K (model-dependent)
* **Duration:** 4s, 8s, 10s, 12s (video models)
* **Audio:** Enable/disable (for models that support it)
#### **Transform Button**
Appears under generated media:
* **Upscale** - Increase resolution
* **Reframe Video** - Change aspect ratio with AI fill
* **Edit** - Open canvas/video editor
* **Regenerate** - Create new version
* **Download** - Save to disk
### Conversation Starters
Appear on new/empty chats:
#### System Starters
* [**Color Grading Assistant**](/conversation-starters/color-grade-assistant) - Technical analysis and LUT generation
* [**Story Cutter Assistant**](/conversation-starters/story-cutter-assistant) - Transcript analysis and paper cuts
* [**Brand Voice Assistant** ](/conversation-starters/brand-voice-assistant)- Brand-consistent content generation
* [**Video Prompter Assistant** ](/conversation-starters/video-prompter-assistant)- Structured prompt creation
* [**Premiere Pro Guru**](/conversation-starters/premiere-pro-guru) - Workflow help and troubleshooting
#### Custom Starters
* Create your own in **Settings → Starters**
* Drag to reorder
* Appear in new chats and are searchable by RAG system
### Sidebar
Access via the sidebar toggle button:
#### Features
* **Conversation List** - All your previous chats
* **Search** - Find conversations by content
* **New Chat** - Start fresh conversation
* **Delete** - Remove conversations
* **Rename** - Edit conversation titles
### [Usage Panel](/getting-started/interface-overview/usage-panel)
Access via **Gear → Usage**:
#### Information Displayed
* **Fal.ai Usage:**
* Last 24 hours
* Last 7 days
* Last 30 days
* **Quick Links:**
* View in Fal Dashboard
* Add Credits
* Billing & Usage
#### Refresh Button
* Updates usage statistics from Fal.ai API
* Shows cached data if offline
### Settings Panel
Access via **Gear → Settings**:
#### Tabs
1. **API Keys**
* Fal.ai API key
2. **Microphone**
* Select input device
* Test microphone
3. **Conversation Starters**
* View system starters
* Create custom starters
* Reorder starters
4. **AI Learning**
* View remembered facts
* Edit/delete memory
* Clear all memory
* Analyze current chat
5. **Prompt System File**
* Upload brand guidelines
* Custom instructions
* Project briefs
**Rather explore it with the founder than figure it out alone?** Book a Founder Onboarding Session — 60 minutes, live inside your actual project. [BOOK YOUR SESSION →](https://checkout.chatvideopro.com/checkout/buy/6b983504-8882-45e4-9ca8-64ea1a28eddf)
***
**Next:** Try the First 5 Things to Try to get started quickly!
# Model Selection
Chat Video Pro has two different model selection systems, each serving a specific purpose. Understanding the difference helps you choose the right workflow for your needs.
### How to Use the Different Model Selectors
{% embed url="" %}
Chat Video Pro has more than one model selector because different jobs need different levels of control.
The easiest way to think about it:
Path
Best for
Top-right selector
Choosing the chat model and default quick image model.
Generate Media selector
Direct control over image and video generation models.
Studio
Guided workflows like Cinematic Lab, Motion Director, AI Transitions, Avatar Studio, Rotoscope, Erase Objects, Add Effects, Reshoot, Upscale, Motion Capture, Multi-Cam, Relight Scene, and Reframe.
If you are not sure where to start, choose based on the job rather than the model name.
***
### Top-Right Selector
The top-right selector lives in the chat header.
It controls:
* The AI model used for chat and text responses.
* The default image model used for quick natural-language image requests.
Use it when:
* You are having a normal conversation with Chat Video Pro.
* You want faster or deeper assistant reasoning.
* You want a quick image without opening Generate Media.
* You type something like `generate an image of a sunset over mountains`.
Do not use it for:
* Video generation.
* Detailed image settings.
* Aspect ratio, duration, resolution, audio, or model-specific controls.
* Studio workflows.
#### Quick Image Example
1. Set your default image model in the top-right selector.
2. Type: `Generate an image of a cozy coffee shop at golden hour.`
3. Chat Video Pro generates the image using default settings.
This is the fastest path, but it is not the most controlled path.
#### Chat Models
Model
Best for
Claude Sonnet 5
Default for chat and Story Cutter. Fast frontier model for everyday editing help, brainstorming, and rough-cut work.
GPT-5.5
Strong general assistant for mixed creative and technical questions.
Gemini 3.1 Pro
Long-context tasks, multimodal reasoning, and Google-native workflows.
Claude Opus 4.8
Complex editorial decisions that need deep reasoning. Available in general chat now; Story Cutter support coming soon.
Claude Fable 5
Prompt writing and creative storytelling. The picker shows a "Very expensive" warning before you send. Story Cutter support coming soon.
Gemini 3.5 Flash
Fast, low-cost second speed tier. Turn it on in Settings when you want a lighter Google model alongside GPT-5.5.
#### Models Settings
You control which models appear in the top-right picker and in Generate Media lists.
1. Open **Settings**.
2. Go to **Configuration → Models**.
3. Search for a model or scroll the list.
4. Toggle a model on to show it, off to hide it.
Four models show by default: **Sonnet 5**, **GPT-5.5**, **Gemini 3.1 Pro**, and **Fable 5**. Enable the rest when you need them. Hidden models stay available in the app; they just do not clutter your picker.
***
### Generate Media Selector
The Generate Media selector appears in the composer when **Generate Media** is enabled.
Use it when:
* You want to generate video.
* You want full control over image generation.
* You need aspect ratio, duration, resolution, quality, or audio settings.
* You are choosing a specific image or video model.
* You are using the classic composer-based generation flow.
Generate Media is the direct model-control path.
#### What It Controls
Depending on your inputs, Generate Media can show:
* Text-to-video models.
* Image-to-video models.
* Transition models.
* Reference models.
* Text-to-image models.
* Image-to-image models.
* Classic video editor tools from an existing video.
Use Supported Video Models and Supported Image Models if you need help choosing a model.
***
### Studio
Studio is not just another model selector. It is the guided workflow layer.
Use Studio when you know the creative task you want to run:
* Create a cinematic still.
* Animate a still with a camera move.
* Build an AI transition.
* Generate alternate angles.
* Remove a background.
* Erase an object.
* Add effects.
* Reshoot a short segment.
* Upscale a clip.
* Relight an image or video.
* Transfer motion to a character image.
Studio chooses or constrains the model path for the workflow, asks for the right source media, and gives you task-specific controls.
Example: instead of choosing a Kling image-to-video model manually and writing a camera prompt, open Motion Director, load a still image, and choose a camera movement preset.
Use Studio when you want a guided production workflow. Use Generate Media when you want direct model control.
***
### Which Path Should I Use?
Goal
Best path
Ask questions, brainstorm, troubleshoot, or get Premiere help
Chat with top-right model selector
Generate a quick image from natural language
Top-right selector
Generate an image with exact aspect ratio or quality settings
Generate Media
Generate video from text
Generate Media
Animate one still image manually
Generate Media Image-to-Video
Animate one still image with camera presets
Studio Motion Director
Connect two frames manually
Generate Media Transition Mode
Connect two frames with guided styles
Studio AI Transitions
Use references for character/product consistency
Generate Media Reference Mode
Create a cinematic still for video work
Studio Cinematic Lab
Clean up, relight, reshoot, add effects, or upscale video
Studio
Edit an existing generated video from chat
Video Canvas Editor or Studio, depending on task
The simple rule:
* Use **Chat** for conversation and quick images.
* Use **Generate Media** for direct model control.
* Use **Studio** for guided creative workflows.
***
### How Attachments Change The Model List
Generate Media adapts to what you attach. This is intentional. It keeps irrelevant models out of the way.
#### No Attachments
Available paths usually include:
* Text-to-Video.
* Text-to-Image.
Use this when you are starting from a prompt.
#### One Image Attached
Available paths usually include:
* Image-to-Video.
* Image-to-Image.
* Reference-capable options, depending on model.
Use this when you want to animate or edit a still image.
Most families keep their text-to-video tier selectable here, treating the image as a reference. **Flux 3 is stricter:** attaching any image commits you to the Flux 3 image-to-video tier, and the Flux 3 text-to-video tier hides. Remove the image to get it back. **Luma Ray 3.2 behaves the same strict way** — with an image attached, only its image-to-video tier is offered.
#### Two Images Attached
Two images often activate Transition Mode, because the app treats them as a start frame and end frame.
Use this when you want one image to become the other.
If you meant to use the two images as subject references instead, switch to Reference Mode with a reference-capable model.
On **Flux 3**, the two-image case is handled by the same image-to-video entry, which routes to a dedicated first/last-frame endpoint behind the scenes. There is no separate Flux 3 transition model to pick.
On **Luma Ray 3.2**, two images are also handled by the image-to-video entry itself — but nothing re-routes: the second image is passed natively as the end frame on the same endpoint. Luma interpolations are 5 seconds only and silent.
The **Seedance** tiers work the Luma way too — the second image is the end frame on the same endpoint, with no separate transition entry. **Seedance 2.5** can carry that interpolation for up to 30 seconds (or Auto), at 480p, 720p, or 1080p with native audio; Seedance 2 covers 4–15 seconds and reaches 1080p and 4K. Unlike Flux 3 and Luma, the Seedance group does not narrow to a single tier when images are attached — at one or two images you see the main tiers and the Reference tiers together.
#### Multiple Reference Images
Multiple images can be used for character, product, subject, or style consistency when you choose a reference-capable model.
Use this when the generated video should keep a person, product, or visual identity consistent across shots.
Not every family has a reference tier. **Flux 3 disappears from the picker once you attach three or more images**, handing off to a reference-capable model such as Veo 3.1 Reference. The same applies when you use Element tags or attach a video clip — Flux 3 has no reference or video-to-video tier. **Luma Ray 3.2 also has no reference tier** and leaves the picker the same way at three or more images or with Element tags — though unlike Flux 3 it does have a video-to-video model, which appears when you attach a clip.
The **Seedance** family does the opposite: at three or more images (or with Element tags, or with a video clip attached) it narrows to its Reference tiers — **Seedance 2 Reference**, **Seedance 2 Fast Reference**, **Seedance 2 Mini Reference**, and **Seedance 2.5 Reference**. Escalation stays inside its own version, so a Seedance 2.5 job that gains a third image lands on Seedance 2.5 Reference, never on a Seedance 2 tier.
#### One Video Attached
When a video is attached, the model list shifts toward video editing and video-to-video options.
Attaching a clip surfaces **Luma Ray 3.2 Edit** in the video-to-video list under the Luma group — a restyle/edit model with a 4-way divergence dial (Default, Adhere, Balanced, Reimagine) that sets how far the result strays from the source. Its output follows the source clip's aspect ratio and carries no generated audio. It is also selectable inside Studio → Add Effects.
For most new video editing tasks, use Studio instead. Studio opens the right workflow directly: Rotoscope, Erase Objects, Add Effects, Reshoot, Upscale, Relight Scene, and more.
***
### Common Workflows
#### Quick Image During Chat
1. Choose your default image model in the top-right selector.
2. Type a natural image request in chat.
3. Use the result or regenerate if needed.
Best for: quick ideas and casual image generation.
#### Full-Control Image Generation
1. Enable **Generate Media**.
2. Choose an image model.
3. Set aspect ratio and quality/resolution.
4. Write the prompt.
5. Generate.
Best for: production images, thumbnails, exact formats, and model comparison.
#### Direct Video Generation
1. Enable **Generate Media**.
2. Choose a video model.
3. Set duration, aspect ratio, resolution, and audio.
4. Write the prompt.
5. Generate.
Best for: text-to-video, image-to-video, transition mode, and reference mode when you want direct model control.
#### Guided Studio Workflow
1. Open **Studio**.
2. Choose the workflow card.
3. Load the required media.
4. Use the workflow controls.
5. Generate and review.
Best for: tasks with a clear shape, like cinematic stills, motion direction, transitions, cleanup, relighting, effects, and upscaling.
***
### Choosing The Right Model
Do not start by memorizing model names. Start with the job.
For video:
* Use Supported Video Models when choosing between Veo, Kling, Sora, Seedance (including 4K, Mini, and Seedance 2.5), Omni Flash, Wan, Hailuo 03, Grok Imagine 1.5, Flux 3, and Luma Ray 3.2.
* Use Text-to-Video, Image-to-Video, Transition Mode, or Reference Mode based on your inputs.
For images:
* Use Supported Image Models when choosing between Nano Banana, GPT Image 2, Flux, Seedream 5.0 Pro, Grok, and Ideogram V4 Fast.
* Use Text-to-Image for normal image generation.
* Use Cinematic Lab when the image is a cinematic frame or source for video.
***
### Troubleshooting
#### I cannot generate videos from the top-right selector
That is expected. The top-right selector is for chat and quick image defaults. Enable **Generate Media** to generate video.
#### The wrong models are showing
Check what is attached to the composer. Attachments change the model list:
* No attachment: text-to-image and text-to-video.
* One image: image-to-image and image-to-video. On Flux 3 the text-to-video tier hides here.
* Two images: transition mode may appear.
* Multiple reference images: reference models may appear. Families without a reference tier, such as Flux 3 and Luma Ray 3.2, drop out of the list entirely.
* Video: video editing or video-to-video options.
#### I attached two images but wanted references
Two images often trigger Transition Mode. Manually switch to a reference-capable model if the images are examples of the same subject rather than start/end frames.
#### I want the app to guide me instead of choosing models
Use Studio. Studio is designed for guided creative tasks and reduces the amount of model selection you need to think about.
#### The settings changed when I switched models
That is normal. Different models support different durations, resolutions, aspect ratios, reference counts, and audio options.
***
### Related Pages
* [Video Generation](/features/video-generation) - Generate video with direct model control.
* [Image Generation](/features/image-generation) - Generate and edit images.
* [Studio](/features/studio) - Use guided production and post-production workflows.
* [Supported Video Models](/features/video-generation/supported-video-models) - Choose a video model.
* [Supported Image Models](/features/image-generation/supported-image-models) - Choose an image model.
***
**Next:** If you want direct video model control, start with Video Generation. If you want a guided workflow, start with Studio.
# Conversation Starters
Conversation Starters are pre-configured assistants that help you get started quickly with specific tasks. They appear on new or empty chats and provide specialized workflows.
## Getting Started
{% embed url="" %}
## Where to Find Them
Conversation Starters appear:
* **On new chats** - When you start a fresh conversation
### System Starters
### Custom Starters
#### Creating Your Own
1. **Gear icon** → **Settings** → **Conversation Starters**
2. Click **"+ Create New Starter"**
3. **Enter:**
* **Title** - Name of your starter (e.g., "Product Review Assistant")
* **Prompt** - Full system prompt that defines the assistant's behavior
4. **Save** - Your starter appears in new chats
#### Managing Starters
* **Reorder:** Drag starters to change their order
* **Edit:** Click on a starter to modify it
* **Delete:** Remove custom starters you no longer need
* **System starters** cannot be deleted, only hidden
#### Best Practices
**Writing effective prompts:**
* Be specific about the assistant's role
* Include examples of desired output
* Define the tone and style
* Mention any constraints or rules
**Example custom starter:**
```
Title: Product Review Assistant
Prompt: You are a product review assistant. Help users create
detailed product review videos. Always include: product name,
key features, pros and cons, and a final recommendation.
Keep reviews honest and balanced.
```
### How Starters Work
#### When You Click a Starter
1. **New conversation starts** with the starter's system prompt
2. **Chat is renamed** to match the starter (e.g., "Color Grading Assistant")
3. **Assistant is configured** with specialized behavior
4. **You can begin** using the assistant immediately
#### Switching Starters
* **Start a new chat** to use a different starter
* **Previous conversations** retain their starter configuration
* **You can't change** the starter of an existing conversation
### Tips for Using Starters
1. **Start with system starters** - They're optimized for common tasks
2. **Create custom starters** for repetitive workflows
3. **Use descriptive titles** - Makes it easy to find the right starter
4. **Iterate on prompts** - Refine custom starters based on results
5. **Share starters** - Copy prompts to share with team members
### Troubleshooting
#### "Starter doesn't appear"
* Check you're on a new or empty chat
* Verify the starter is enabled in Settings
* Try refreshing the panel
#### "Starter behavior is wrong"
* Edit the starter prompt in Settings
* Be more specific about desired behavior
* Include examples in the prompt
#### "Can't create custom starter"
* Check you're in Settings → Conversation Starters
* Ensure you've filled in both title and prompt
* Try saving again
***
**Next:** Learn about the[ Generate Media Button](/getting-started/interface-overview/generate-media-button) to start creating AI-powered content.
# Generate Media Button
The Generate Media button toggles between chat mode and media generation mode, giving you access to AI-powered video and image generation tools.
{% embed url="" %}
### Where to Find It
The Generate Media button is located in the composer toolbar, near the other input tools.
Use it when you want to create or edit media directly from the composer instead of opening a guided Studio workflow.
***
### What It Does
Generate Media changes the composer from normal chat into a media generation workspace.
#### Chat Mode
When Generate Media is off:
* The composer works like a normal chat box.
* You can ask questions, plan edits, troubleshoot, and use conversation starters.
* File uploads still work.
* Chat Video Pro responds with text, analysis, instructions, or tool outputs depending on the request.
Use Chat Mode for:
* Premiere Pro questions.
* Prompt brainstorming.
* Story Cutter, Color Grade Assistant, Brand Voice Assistant, and Premiere Pro Guru.
* General help and planning.
#### Generate Media Mode
When Generate Media is on:
* The model selector appears.
* Media settings appear.
* The composer expects a generation prompt.
* Available models adapt to your attachments.
Use Generate Media for:
* Text-to-Video.
* Image-to-Video.
* Text-to-Image.
* Image-to-Image.
* Transition Mode.
* Reference Mode.
***
### Generate Media vs. Studio
Generate Media and Studio both create media, but they are designed for different moments.
Use Generate Media when...
Use Studio when...
You want direct model control.
You want a guided workflow.
You already know the model and settings you want.
You want the tool to choose the right structure for the task.
You are making a prompt-based image or video.
You are doing a named workflow like Cinematic Lab, Motion Director, AI Transitions, Rotoscope, Erase Objects, Add Effects, Reshoot, Upscale, Multi-Cam, Motion Capture, or Relight Scene.
You want to compare models manually.
You want workflow-specific controls, presets, asset loading, and review steps.
The simple rule: use **Generate Media** for direct model control and **Studio** for guided creative workflows.
See Studio.
***
### How To Use Generate Media
#### Enabling Generate Media
1. Click the **Generate Media** toggle in the composer.
2. Media controls appear above the text input.
3. Select your model.
4. Adjust available settings.
5. Enter your prompt.
6. Generate.
#### Disabling Generate Media
1. Click the toggle again.
2. Media controls hide.
3. The composer returns to normal chat.
If you are asking a question instead of generating media, turn Generate Media off.
***
### Media Settings
When Generate Media is enabled, you will see model and settings controls.
#### Model Selector
The model selector shows compatible models based on what you have attached.
Common groups include:
* **Video models:** Sora, Veo, Seedance, Kling, Hailuo 03, Wan, Grok Imagine 1.5, and others.
* **Image models:** Nano Banana, GPT Image, Flux, Seedream 5.0 Pro, Grok, Ideogram V4 Fast, and others.
* **Video editing models:** Rotoscope, VFX, reshoot, upscaling, and related tools when video is attached.
If the model you expect is missing, check your attachments. The composer may be in a different mode because it sees an image, two images, references, or a video.
#### Aspect Ratio
Aspect ratio controls the shape of the result.
Common choices:
* **16:9** - YouTube, landscape video, standard editing timelines.
* **9:16** - TikTok, Instagram Reels, YouTube Shorts, vertical ads.
* **1:1** - Square social posts.
* **4:3** - Classic, vintage, or presentation-style framing.
Click the expanded menu when available to see more model-specific shapes.
Choose the final delivery shape before generating. Reframing later is possible, but generations are usually stronger when the model starts in the right format.
#### Resolution and Quality
Resolution options vary by model.
Use lower or standard settings for early drafts, then use higher quality settings after the prompt and direction are working.
Upscale final images or videos after the creative result is approved.
#### Duration
Duration appears for video models.
Shorter clips are easier to control and cheaper to test. Start short, confirm the direction, then generate longer clips only when the action needs more time.
#### Audio
Some video models support generated audio.
Enable audio when:
* The scene benefits from ambience, dialogue, sound effects, or natural environment sound.
* The selected model supports audio.
* You are willing to evaluate both picture and sound.
Disable audio when:
* You plan to design sound in Premiere later.
* You want to reduce variables while testing.
* The model or mode does not support audio.
***
### Automatic Mode Detection
Generate Media adapts based on your attachments.
#### No Attachments: Text-to-Video or Text-to-Image
With no images or videos attached, the composer is ready for prompt-only generation.
Use this when you are starting from an idea.
#### One Image Attached: Image-to-Video or Image-to-Image
With one image attached, image-based models appear.
Use this when you want to:
* Animate a still image.
* Edit an image.
* Use a captured frame as source material.
* Create a variation from an existing image.
If the goal is a controlled camera move from one still, consider Studio Motion Director.
#### Two Images Attached: Transition Mode
With two images attached, Chat Video Pro can treat them as a start frame and end frame.
Use this when you want one image to become another.
For guided transition styles, use Studio AI Transitions.
#### Multiple Reference Images: Reference Mode
With multiple reference images and a reference-capable model, you can guide identity, product consistency, style, wardrobe, or environment.
Use references when consistency matters more than a one-off prompt.
See Reference Mode.
#### One Video Attached: Video Editing or Video-to-Video
With one video attached, video editing models and video-to-video options can appear.
For most named video editing jobs, Studio is the clearest path:
* Rotoscope
* Erase Objects
* Add Effects
* Reshoot
* Upscale
* Relight Scene
Use the classic video editor when you are already working from a generated/imported clip and want a quick direct edit. Use Studio when you want the guided version of the workflow.
***
### Workflow Examples
#### Example 1: Generate a Video
1. Enable **Generate Media**.
2. Select a video model.
3. Set aspect ratio, duration, resolution, and audio options.
4. Enter a prompt:
{% code overflow="wrap" %}
```
A cinematic shot of a mountain road at sunrise, slow drone push forward, warm golden light, mist in the valley, realistic documentary style.
```
{% endcode %}
5. Generate.
6. Review and revise.
#### Example 2: Create an Image
1. Enable **Generate Media**.
2. Select an image model.
3. Set aspect ratio and quality/resolution.
4. Enter a prompt:
{% code overflow="wrap" %}
```
A modern coffee shop interior, warm morning light, natural wood, plants, cinematic editorial photography.
```
{% endcode %}
5. Generate.
6. Reuse the result as a thumbnail base, reference image, or source frame.
#### Example 3: Animate an Image
1. Attach an image.
2. Enable **Generate Media**.
3. Choose an image-to-video model.
4. Enter a motion prompt:
{% code overflow="wrap" %}
```
Slow push-in with subtle parallax, preserve the subject and lighting, cinematic natural movement.
```
{% endcode %}
5. Generate.
For preset camera movement, use Studio Motion Director.
***
### Tips for Best Results
* Choose the destination aspect ratio before generating.
* Use one main creative idea per generation.
* Keep video tests short until the direction works.
* Use reference images when identity, product, or style consistency matters.
* Use Studio for structured workflows instead of forcing everything through one prompt.
* Upscale only after the creative result is approved.
* If a model disappears, remove extra attachments and try again.
***
### Troubleshooting
#### Generate Media button does not appear
* Check that you are in the composer area.
* Wait for the panel to finish loading.
* Restart Premiere Pro if the panel is stuck.
#### Media settings do not show
* Make sure Generate Media is enabled.
* Confirm your API key is saved.
* Confirm your provider account has credits.
* Try removing attachments and turning Generate Media on again.
#### Model selector is empty or missing the model I want
* Check your fal.ai account balance.
* Verify the API key is correct.
* Remove attachments that may be forcing the composer into another mode.
* Try a supported input type for the model you want.
* Check whether the provider model is temporarily unavailable.
#### Generation fails
* Check your fal.ai balance.
* Verify your internet connection.
* Try a shorter prompt or smaller request.
* Check model-specific requirements.
* Use Generation Errors & Failed Jobs FAQ.
***
**Next:** Learn about the Frame Capture Button or use Studio when you want a guided workflow.
# Frame Capture Button
### Where to Find It
The Frame Capture button is located in the **Composer toolbar**, to the left of the text input area. It appears as a camera icon (📷).
### What It Does
When you click the Frame Capture button:
1. **Captures the current frame** from your Premiere Pro timeline at the playhead position
2. **Exports it as an image** (JPEG format for optimal storage)
3. **Adds it to the composer** as an attachment
4. **Ready to use** for color grading, analysis, or as reference for generation
### How to Use
#### Basic Workflow
1. **Position your playhead** in Premiere Pro on the frame you want to capture
2. **Click the Frame Capture button** (📷) in Chat Video Pro
3. **Wait for export** - Button shows a pulse animation while processing
4. **Frame appears** in the composer as an image attachment
5. **Use the frame** for:
* Color grading analysis
* Asking questions about the shot
* Using as reference for generation
* Creating matching looks
#### Color Grading Workflow
1. **Capture frame** using Frame Capture button
2. **Open Color Grading Assistant** (or ask "help me with color grading")
3. **Ask for analysis:** "What's wrong with this shot?" or "Analyze the color"
4. **Get recommendations** with one-click Apply button
5. **Request creative LUT:** "Create a cinematic LUT" or "Give me a Blade Runner look"
#### Reference Workflow (from within the Color Grade Assistant)
1. **Capture frame** from your timeline
2. **Use as reference** for:
* Matching colors in generated content
* Creating consistent looks
* Style transfer
* Image-to-video with specific aesthetic
### Technical Details
#### Export Format
* **Format:** JPEG (converted from PNG for better storage efficiency)
* **Quality:** High quality, suitable for analysis
* **Size:** Optimized for chat storage and display
#### Frame Accuracy
* **Exact frame** at playhead position
* **No interpolation** - Captures the actual frame shown
* **Frame-accurate** - Matches what you see in Premiere Pro
#### Requirements
* **Active sequence** - Must have a sequence open in Premiere Pro
* **Playhead positioned** - Must be on a valid frame
* **Sufficient disk space** - Temporary files are created during export
### Use Cases
#### Color Grading
* Capture frame → Get technical analysis → Apply corrections
* Capture frame → Request creative LUT → Download and apply
#### Shot Analysis
* Ask questions about composition, lighting, or technical issues
* Get recommendations for improvements
* Understand what's working and what needs adjustment
#### Reference for Generation
* Add, remove, or change aspects of the image with image-to-image generation
* Generate thumbnails for your videos
* Animate the image with image-to-video
### Tips for Best Results
1. **Capture high-quality frames** - Ensure your sequence is at full quality
2. **Position playhead carefully** - Capture the exact frame you want
3. **Use for color grading** - Frame Capture is optimized for color analysis
4. **Combine with Color Grading Assistant** - Best workflow for professional grading
5. **Multiple frames** - Capture multiple frames for sequence-wide analysis
### Troubleshooting
#### "Frame export failed"
**Common causes:**
* No active sequence in Premiere Pro
* Playhead not positioned on a valid frame
* Insufficient disk space for temporary files
**Solutions:**
* Ensure a sequence is open and active
* Position playhead on a frame with video content
* Check available disk space
* Try capturing a different frame
#### "Frame doesn't appear in composer"
* Check the export completed (button stops pulsing)
* Verify the image was added to attachments
* Try capturing again
* Check browser console for errors (if available)
#### "Frame quality is low"
* Ensure sequence is at full quality (not proxy mode)
* Check source footage quality
* Frame is optimized for analysis, not final export
* For final exports, use Premiere Pro's export functions
#### "Export is slow"
* Large sequences may take longer
* Complex effects can slow export
* First export may be slower (initialization)
* Subsequent exports are typically faster
### Keyboard Shortcuts
* **No direct shortcut** - Click the button in the composer
### Related Features
* [**Clip Export Button**](/getting-started/interface-overview/import-clip-button) - Export entire clips (not just frames)
* [**Color Grading Assistant**](/conversation-starters/color-grade-assistant) - Analyze captured frames
* [**LUT Generation**](/conversation-starters/color-grade-assistant) - Create looks from captured frames
* [**Style Transfer**](/conversation-starters/color-grade-assistant) - Match colors from captured frames
***
**Next:** Learn about the [Import Clip Button](/getting-started/interface-overview/import-clip-button) to bring entire clips from your timeline into Chat Video Pro.
# Import Clip Button
The Import Clip button (🎥) exports the selected clip from your Premiere Pro timeline directly into Chat Video Pro for video-to-video editing, SAM 3 rotoscoping, or AI processing.
## Where to Find It
The Import Clip button is located in the **Composer toolbar**, to the left of the Frame Capture button. It appears as a video camera icon (🎥).
### What It Does
When you click the Import Clip button:
1. **Detects the selected clip** on your Premiere Pro timeline
2. **Exports the clip region** (respects in/out points if set)
3. **Trims to timeline selection** (not the entire source file)
4. **Generates a thumbnail** for preview
5. **Adds to composer** as a video attachment
6. **Ready for editing** with SAM 3, Kling VFX, or other video tools
### Important: Source Video vs. Rendered Effects
#### Understanding What Gets Imported
**Regular Clip Export (Default Behavior):**
* ✅ **Source video file** - The original media file
* ✅ **Timeline in/out points** - Respects your selection
* ❌ **No color grading** - Lumetri corrections are NOT included
* ❌ **No effects** - Time remapping, effects, titles are NOT included
* ❌ **No adjustments** - Adjustment layers are NOT included
**Why this matters:**
* If you've applied color grading, effects, or time remapping in Premiere Pro, these won't be included in the exported clip
* The imported video will be the raw source file, not the rendered version you see on your timeline
#### Including Color Grading and Effects
**To include all your Premiere Pro work (color, effects, time remapping):**
1. **Nest the clip first** in Premiere Pro:
* Select your clip(s) on the timeline
* Right-click → **Nest** (or use keyboard shortcut)
* Give the nested sequence a name
* The nested sequence now contains all your effects, color, and adjustments
2. **Select the nested sequence** on your timeline
3. **Click Import Clip button** in Chat Video Pro
4. **Media Encoder export** - Chat Video Pro automatically uses Adobe Media Encoder to export the nested sequence
* Includes all color grading (Lumetri)
* Includes all effects and adjustments
* Includes time remapping
* Matches your sequence settings (frame rate, resolution)
* Takes a few seconds longer (Media Encoder processing)
**Workflow comparison:**
| Method | Includes Color/Effects | Speed | Use When |
| -------------------------- | ---------------------- | ------ | -------------------------------------------- |
| **Direct clip export** | ❌ No | Fast | Raw footage, no effects needed |
| **Nested sequence export** | ✅ Yes | Slower | Color graded, effects applied, time remapped |
### How to Import Videos
#### Method 1: Import Clip Button (Timeline → Composer)
**Best for:** Clips already on your Premiere Pro timeline
1. **Select a clip** on your Premiere Pro timeline (click on it)
2. **Click the Import Clip button** (🎥) in Chat Video Pro
3. **Wait for export** - Button shows pulse animation while processing
4. **Clip appears** in the composer as a video thumbnail
**For clips with effects:**
1. **Nest the clip first** (Right-click → Nest)
2. **Select the nested sequence**
3. **Click the Import Clip button**
4. **Wait for Media Encoder export** (takes longer but includes everything)
#### Method 2: Drag & Drop from Library
**Best for:** Previously generated videos or saved media
1. **Open Library** - Click Library icon in sidebar or use Library panel
2. **Find your video** - Browse generated videos, imported clips, etc.
3. **Drag video** from Library
4. **Drop onto the composer** - Video appears as an attachment
5. **Ready to edit** - Same as imported clips
**Library videos include:**
* Previously generated videos
* Videos imported from the timeline
* Videos saved to Library
* All your Chat Video Pro media
#### Method 3: Drag & Drop from File System
**Best for:** External video files
1. **Open file explorer/finder**
2. **Find your video file** (MP4, MOV, etc.)
3. **Drag the video file** onto the Chat Video Pro composer
4. **Video appears** as an attachment
5. **Ready to edit**
### Technical Details
#### What Gets Imported (Regular Clips)
* **Source media file** - Original video file from disk
* **Selected region** - Only the portion on your timeline
* **Timeline in/out points** - Respects your selection
* **No rendering** - Uses source file directly (fast)
#### What Gets Imported (Nested Sequences)
* **Rendered video** - Fully processed with all effects
* **Color grading included** - All Lumetri corrections
* **Effects included** - Time remapping, effects, titles
* **Adjustment layers** - All adjustments rendered
* **Sequence settings** - Matches your sequence frame rate and resolution
* **Media Encoder export** - Uses Adobe Media Encoder (slower)
#### Supported Clip Types
**Regular Clips:**
* Imported instantly using source media file
* Fast export, no processing needed
* No effects/color included
**Nested Sequences:**
* Automatically exported via Adobe Media Encoder
* Takes a few seconds longer
* Includes all effects and color grading
* Matches sequence settings
**Adjustment Layers, Titles, Graphics:**
* Require nesting first (to include in export)
* Processed through Media Encoder
* May take longer depending on complexity
#### Export Limits
* **Maximum file size:** 1 GB
* **Maximum duration:** 5 minutes
* **Longer clips:** Trim on timeline before exporting
* **Nested sequences:** May take longer depending on effects complexity
#### File Format
**Regular Clips:**
* Uses source file format directly
* No conversion needed
**Nested Sequences:**
* **Export format:** H.264 MP4 (for compatibility)
* **Quality:** Matches timeline quality settings
* **Frame rate:** Matches sequence frame rate
* **Resolution:** Matches sequence resolution
### Tips for Best Results
1. **Nest first for effects** - If you want color/effects included, nest the clip before exporting
2. **Keep clips short** - Under 30 seconds for faster SAM 3 processing
3. **Trim on the timeline first** - Only export what you need
4. **Select single clips** - Import one clip at a time for best results
5. **Check file size** - Keep under 1 GB limit
6. **Use Library for reuse** - Drag videos from Library to reuse in multiple projects
### Troubleshooting
#### "No clip selected"
**Solutions:**
* Click directly on a clip in your timeline
* Ensure the clip is actually selected (highlighted)
* Check that you have an active sequence
* Try selecting a different clip
#### "Color grading not included"
**Solution:** You need to nest the clip first:
1. Select your clip(s) on the timeline
2. Right-click → Nest
3. Select the nested sequence
4. Click the Import Clip button
5. Media Encoder will export with all effects included
#### "Clip export taking too long"
**Common causes:**
* Nested sequence (requires Media Encoder export)
* Adjustment layers or graphics (need rendering)
* Very long clip (trim on timeline first)
* Complex effects (may slow export)
**Solutions:**
* Wait for export to complete (shows progress)
* Trim the clip to a shorter duration on the timeline
* Simplify nested sequences if possible
* Check Media Encoder is running (for nested sequences)
#### "Export failed"
**Common causes:**
* Clip is offline or missing source file
* Insufficient disk space
* File format not supported
* Clip duration exceeds 5-minute limit
**Solutions:**
* Relink offline media in Premiere Pro
* Check available disk space
* Ensure the source file is a valid video format
* Trim clip to under 5 minutes
#### "File size too large"
* Trim the clip to a shorter duration
* Use lower quality export settings (if available)
* Split into multiple shorter clips
* Check source file size
### Related Features
* [**Frame Capture Button** ](/getting-started/interface-overview/frame-capture-button)- Export single frames (not entire clips)
* [**Video Canvas Editor** ](/features/video-generation/video-canvas-editor)- Edit imported clips
* [**SAM 3**](/features/studio/sam-3-rotoscoping) - Professional rotoscoping tool
* [**Kling VFX** ](/features/studio/kling-vfx)- Visual effects and scene modifications
* [**Library**](/getting-started/interface-overview/library) - Store and reuse imported videos
# Voice Input
Voice Input allows you to speak your prompts instead of typing, making Chat Video Pro hands-free and faster to use.
## Where to Find It
The Voice Input button is located in the **Composer toolbar**, typically to the left of the send button. It appears as a microphone icon (🎤).
### What It Does
When you click the Voice Input button:
1. **Starts recording** from your selected microphone
2. **Transcribes speech** to text in real-time (when supported)
3. **Stops recording** when you click again or after a timeout
4. **Inserts transcribed text** into the composer
5. **Ready to send** - Review and edit if needed, then send
### How to Use
#### Basic Workflow
1. **Click the Voice Input button** (🎤) and wait for it to pulse red
2. **Speak your prompt** clearly
3. **Click again to stop** recording
4. **Review the transcribed text** in the composer
5. **Edit if needed** (add details, fix transcription errors)
6. **Send** your message
#### Setting Up Your Microphone
1. **Gear icon** → **Settings** → **Microphone**
2. **Select your input device** from the dropdown
3. **Test the microphone** using the test button
4. **Adjust volume** if needed
5. **Save settings**
#### Best Practices
1. **Speak clearly** - Enunciate words for better transcription
2. **Minimize background noise** - Use a quiet environment
3. **Speak at normal pace** - Not too fast or too slow
4. **Review transcription** - Check for errors before sending
5. **Edit as needed** - Add details or fix mistakes
### Voice Input States
#### Idle State
* **Microphone icon** (🎤) - Ready to record
* **Click to start** recording
#### Recording State
* **The button will pulse**
* **Click again to stop** recording
#### Processing State
* **Shows processing indicator** - Transcribing audio
* **Wait for completion** - Text appears when ready
#### Error State
* **Shows an error message** - If recording fails or no dialog is detected
* **Check microphone** settings and permissions
### Use Cases
#### Quick Prompts
**Perfect for:**
* Fast iterations
* Quick questions
* Simple requests
* When typing is inconvenient
#### Hands-Free Operation
**Great when:**
* Your hands are busy (adjusting Premiere Pro)
* You're describing what you see
* You want to speak naturally
* You're working with a team (can speak while others watch)
#### Long Prompts
**Useful for:**
* Detailed descriptions
* Complex requests
* Natural language explanations
* When typing would be slow
### Technical Details
#### Microphone Requirements
* **Any USB or built-in microphone**
* **System default** or manually selected device
* **Permissions** - Browser must have microphone access
### Tips for Best Results
1. **Use a good microphone** - Better hardware = better transcription
2. **Speak clearly** - Enunciate, don't mumble
3. **Minimize noise** - A quiet environment works best
4. **Review before sending** - Always check transcription
5. **Edit as needed** - Add details or fix errors
6. **Practice** - Gets easier with use
### Troubleshooting
#### "Microphone not working."
**Solutions:**
* Check that the microphone is connected and powered
* Verify microphone permissions in browser/OS
* Select the correct device in Settings → Microphone
* Test the microphone using the test button
* Check system audio settings
#### "Transcription is inaccurate."
**Solutions:**
* Speak more clearly and slowly
* Reduce background noise
* Use a better microphone
* Edit transcription manually
* Check microphone volume levels
#### "Voice input button doesn't appear."
**Solutions:**
* Check the browser supports the Web Speech API
* Verify you're in a supported environment
* Try refreshing the panel
* Check for extension updates
#### "Recording stops immediately."
**Solutions:**
* Check microphone permissions
* Verify the microphone is working in other apps
* Try a different microphone
* Check system audio settings
### Related Features
* **Text Input** - Type prompts manually
* **Microphone Settings** - Configure input device
* **Composer** - Where transcribed text appears
***
**Next:** Learn about [Web Search](/getting-started/interface-overview/web-search) to get current information and answers.
# Web Search
Web Search enables Chat Video Pro to access current information from the internet, providing up-to-date answers and real-time data.
### What It Does
Web Search allows the AI assistant to:
* **Search the web** for current information
* **Access recent data,** not in training data
* **Find current events** and news
* **Get up-to-date documentation** and guides
* **Answer questions** requiring real-time information
### How to Enable
#### Requirements
* **Internet connection** - Active connection required
### How to Use
#### Automatic Web Search
The AI automatically uses web search when:
* **Question requires current data** - Recent events, current prices, etc.
* **Training data is outdated** - Information that changes frequently
* **Specific websites needed** - Links to current documentation
* **Real-time information** - Stock prices, weather, news, etc.
#### Manual Web Search
You can explicitly request web search:
**Examples:**
* "Search the web for Premiere Pro 2024 new features"
* "Find current information about \[topic]"
* "Look up the latest \[subject]"
* "Use web search to find \[information]"
#### Using /web Command
Type `/web` followed by your query:
* `/web Premiere Pro crash fixes`
* `/web latest video editing trends`
* `/web Adobe Premiere Pro system requirements`
### When Web Search is Used
#### Automatically Triggered
The AI uses web search when:
* **Current information needed** - Not available in training data
* **Recent events** - Happened after training cutoff
* **Specific documentation** - Links to official sources
* **Real-time data** - Prices, availability, status
#### Suppressed
Web search is **not used** when:
* **Training docs have the answer** - the RAG system provides the information
* **Transcripts contain info** - Uploaded files answer the question
* **General knowledge** - Well-established facts
* **User explicitly asks** not to search
### Use Cases
#### Premiere Pro Help
**Example:**
* "What are the new features in Premiere Pro 2024?"
* AI searches for current release notes and features
* Provides links to official Adobe documentation
#### Technical Troubleshooting
**Example:**
* "How do I fix Premiere Pro crash on Windows 11?"
* AI searches for current solutions and forum posts
* Provides up-to-date troubleshooting steps
#### Current Events
**Example:**
* "What are the latest AI video generation models?"
* AI searches for recent announcements and releases
* Provides current information about new models
#### Documentation Links
**Example:**
* "Where can I find Fal.ai API documentation?"
* AI searches and provides direct links.
* Ensures links are current and working
### Tips for Best Results
1. **Be specific** - More specific queries get better results
2. **Request explicitly** - Say "search the web" if you want current info
3. **Check links** - AI provides sources, verify if needed
4. **Combine with chat** - Ask follow-up questions after search
5. **Use for current data** - Best for information that changes
### Limitations
#### What Web Search Can't Do
* **Access paid content** - Can't bypass paywalls
* **Private information** - Can't access private accounts
* **Real-time streaming** - Not for live data streams
* **Interactive sites** - Can't interact with web apps
#### Accuracy Considerations
* **Verify important information** - Check sources provided
* **Multiple sources** - AI may cite different sources
* **May be outdated** - Some information changes quickly
* **Use judgment** - Verify critical information independently
### Troubleshooting
#### "Web search not working."
**Solutions:**
* Checkthat you have an OpenAI or Gemini API key configured
* Verify the internet connection is active
* Try explicitly requesting a web search
* Check that the API key is valid and has credits
#### "No search results"
**Solutions:**
* Try rephrasing your query
* Be more specific in your request
* Check if information is available online
* Verify API key permissions
#### "Search results are outdated."
**Solutions:**
* Request "latest" or "current" information
* Specify a time frame ("recent", "2024", etc.)
* Check the sources provided by AI
* Verify information independently
### Privacy & Security
* **Search terms** may be logged by the API provider
* **No personal data** should be included in searches
* **Review the privacy policies** of API providers
### Related Features
* **Premiere Pro Guru** - Automatically uses web search for help
* **Chat Assistant** - Can use web search when needed
* **RAG System** - Provides local knowledge (doesn't require web)
***
**Next:** Learn about the [Usage Panel](/getting-started/interface-overview/usage-panel) to track your spending and manage your account.
# Library
The Library is your central hub for all generated media. Access all your images, videos, and LUTs in one place. Reuse content, create reusable elements, and quickly return to where content was created
### What the Library Is
The Library stores and organizes all your generated content:
* **Images** - All generated images from any chat
* **Videos** - All generated videos from any chat
* **LUTs** - 280 free CVP color grades plus LUTs you create with the Color Grade Assistant
* **Elements** - Reusable image collections for consistent character/product references
**Think of it as:** Your personal media library that remembers everything you've created, making it easy to find, reuse, and organize your work.
### Accessing the Library
**How to open:**
* Click the **Library** button in the sidebar
**Library tabs:**
* **Media** - All images and videos
* **LUTs** - CVP LUT Gallery collection and custom LUTs
* **Elements** - Reusable reference collections
### Drag and Drop from Library
You can drag any image or video from the Library back into the composer to continue working with it.
#### How It Works
1. **Open Library** - Click the Library button
2. **Find your media** - Browse or search for the image/video
3. **Drag to composer** - Click and drag the thumbnail
4. **Drop in composer** - Release over the composer area
5. **Continue working** - Use it for new generations or edits
#### Use Cases
**Re-using generated images:**
* Generate an image
* Later, drag it from the Library to the composer
* Use it as a reference for image-to-image editing
* Or use it as input for image-to-video
**Continuing video work:**
* Generate a video
* Later, drag it from the Library to the composer
* Apply effects, upscale, or edit further
**Building on previous work:**
* Find a previous generation
* Drag it back into the composer
* Build variations or continue the creative process
### Elements Feature
Elements are reusable collections of 1-4 images that you can reference in prompts using `@ElementName`. Perfect for maintaining consistency across multiple generations.
#### What Elements Are
Elements are named collections of reference images:
* **Characters** - People, animals, or subjects you want to reuse
* **Objects** - Products, props, or items for consistent appearance
* **Environments** - Backgrounds, locations, or settings
* **Custom** - Any other reusable reference collection
**Why use Elements:**
* **Save time** - No need to re-upload the same images repeatedly
* **Consistency** - Maintain character/product appearance across generations
* **Organization** - Keep your references organized by category
* **Universal** - Works with all models that support image inputs
#### Creating Elements
1. **Open Library** - Click the Library button
2. **Go to the Elements tab** - Click "Elements" tab
3. **Click "+" button** - Create new element
4. **Name your element** - Give it a clear name (e.g., "Raia", "Product Shot")
5. **Select category** - Choose Character, Object, Environment, or Custom
6. **Add 1-4 images** - Upload or select from library
* **Main View** (required) - Primary reference image
* **Angle 2-4** (optional) - Additional views/angles
7. **Save** - Element is now available for use
#### Using Elements in Prompts
Once created, elements are automatically available when you type `@` in prompts.
**How to use:**
1. Start typing a prompt
2. Type `@` to see autocomplete
3. Select your element (e.g., `@Raia`, `@Product`)
4. Continue with your prompt.
**Example prompts:**
```
Create a video of @Raia walking through a forest
```
Generate an image of @Product on a white background
```
Use @Character1 and @Character2 in a conversation scene
```
#### Element Categories
**Character:**
* People, animals, anthropomorphic subjects
* Use for: Consistent character appearance
* Example: `@Raia`, `@MainCharacter`
**Object:**
* Products, props, physical items
* Use for: Product shots, consistent objects
* Example: `@Product`, `@Logo`
**Environment:**
* Locations, backgrounds, settings
* Use for: Consistent scene backgrounds
* Example: `@ForestScene`, `@Office`
**Custom:**
* Any other reference type
* Use for: Specialized references
* Example: `@StyleReference`, `@ColorPalette`
#### Managing Elements
**Edit element:**
* Click on the element in the Library
* Modify name, category, or images
* Save changes
**Delete element:**
* Click on the element
* Click the delete button
* Confirm deletion
**View usage:**
* Elements show usage count
* Most used elements appear first (when sorted by "Most used")
### Downloading Media
You can download any image or video from the Library.
#### How to Download
1. **Open Library** - Click the Library button
2. **Find your media** - Browse or search
3. **Click on media** - Opens detail view
4. **Click Download** - Downloads to your computer
5. **Choose location** - Select where to save
#### Download Options
**Images:**
* Downloads as PNG or original format
* Preserves quality
* Includes metadata if available
**Videos:**
* Downloads as MP4
* Original quality
* Includes audio if present
### Returning to Chats
You can quickly return to the chat where any media was generated.
#### How It Works
1. **Open Library** - Click the Library button
2. **Find your media** - Browse or search
3. **Click on media** - Opens detail view
4. **Click "Return to Chat"** - Opens the original chat
5. **Auto-scrolls** - Automatically scrolls to the message where it was generated
#### Use Cases
**Finding context:**
* See the prompt that created the media
* Review the conversation context
* Understand how it was generated
**Continuing work:**
* Return to the original chat
* Generate variations
* Build on previous work
**Reviewing history:**
* See all messages in that chat
* Review the full workflow
* Understand the creative process
### CVP LUT Gallery
Every Chat Video Pro install ships with **280 free CVP LUTs**. You browse them in the **LUT Gallery**, preview grades on your footage, and apply a look in one click.
**Open the Gallery:**
* **Library → LUTs tab** — browse the full collection alongside LUTs you created
* **Studio** — open LUT Gallery when you want to grade while you work
**What you can do:**
* Preview any grade before you commit
* Apply a LUT to a clip directly from the Gallery
* Organize LUTs into **folders** by project, mood, or client
The Color Grade Assistant **creates** custom LUTs from a conversation. The LUT Gallery **applies** the built-in CVP collection. Use both when you want a starting grade from the library and a bespoke look from the Assistant.
#### Apply a CVP LUT to a clip
1. Open **Library → LUTs** or **LUT Gallery** from Studio
2. Browse or search the CVP collection
3. Preview grades on your clip
4. Click **Apply** on the grade you want
### LUT Copy Path Feature
For LUT files, you can quickly copy the file path and paste it directly into Lumetri Color in Premiere Pro.
#### How It Works
1. **Open Library** - Click the Library button
2. **Go to LUTs tab** - Click "LUTs" tab
3. **Find your LUT** - Browse or search
4. **Click on LUT** - Opens detail view
5. **Save LUT** (if not saved) - Saves to disk first
6. **Click "Copy Path"** - Copies file path to clipboard
7. **Paste in Lumetri** - Paste path into Lumetri Color panel
#### Using Lumetri
**Method 1: Direct path paste**
1. Copy path from Library
2. Open the Lumetri Color panel in Premiere Pro
3. Click on the LUT dropdown
4. Paste the path or navigate to the file location
**Method 2: Browse to the file**
1. Copy path from Library
2. Open the Lumetri Color panel
3. Click "Browse" or "Load LUT"
4. Navigate to the copied path location
### Library Features
#### Search
**How to search:**
* Type in the search box at the top of the Library
* Searches prompts, model names, and filenames
* Works across all tabs (Media, LUTs, Elements)
**Search tips:**
* Search by prompt keywords
* Search by model name
* Search by element name (Elements tab)
#### Filtering
**Media tab filters:**
* **All media** - Everything
* **Images only** - Just images
* **Videos only** - Just videos
* **Favorites** - Starred items only
**Elements tab filters:**
* **All elements** - Everything
* **Characters** - Character elements only
* **Objects** - Object elements only
* **Environments** - Environment elements only
* **Custom** - Custom elements only
#### Sorting
**Sort options:**
* **Most recent** - Newest first (default)
* **Oldest first** - Oldest first
* **Most used** - For elements, shows the most frequently used first
#### Favorites
**How to favorite:**
* Click the star icon on any media item
* Starred items appear in the "Favorites" filter
* Quick access to your best work
### Tips for Best Results
1. **Name elements clearly** - Use descriptive names for easy finding
2. **Use categories** - Organize elements by category for better management
3. **Add multiple angles** - Include 2-4 images for better consistency
4. **Favorite your best work** - Star items you want to reuse
5. **Search regularly** - Use the search to quickly find what you need
6. **Return to chats** - Use "Return to Chat" to see the original context
7. **Copy LUT paths** - Use the copy path for quick Lumetri integration
### Common Workflows
#### Creating a Reusable Character Element
1. Generate images of your character
2. Open Library → Elements tab
3. Click "+" to create element
4. Name it (e.g., "Raia")
5. Select "Character" category
6. Add 2-4 images (different angles/views)
7. Save element
8. Use `@Raia` in future prompts
#### Reusing Previous Work
1. Open Library
2. Search for the previous generation
3. Drag the image/video to the composer
4. Continue editing or generating
5. Build on previous work
#### Quick LUT Application
1. Generate LUT with Color Grade Assistant
2. Open Library → LUTs tab
3. Click on LUT
4. Click "Copy Path"
5. Open Lumetri Color in Premiere Pro
6. Paste path or browse to file
7. Apply LUT instantly
#### Finding Context
1. See generated media in chat
2. Later, find it in the Library
3. Click "Return to Chat"
4. Review original prompt and context
5. Generate variations or continue work
### Troubleshooting
#### "Can't find my media."
**Solutions:**
* Use the search to find it
* Check different tabs (Media, LUTs, Elements)
* Try different filters
* Media is stored per chat - check all chats
* If you cannot find your media, all generations are located at
#### "Element not showing in @ mentions."
**Solutions:**
* Ensure the element is saved
* Check element name (case-sensitive)
* Type `@` to see autocomplete
* Refresh or restart if needed
#### "Can't drag from Library."
**Solutions:**
* Ensure the library is open
* Click and hold on the thumbnail
* Drag to the composer area
* Try clicking on media first, then drag from the detail view
#### "LUT path not copying."
**Solutions:**
* Save LUT first (if not already saved)
* Click the "Copy Path" button
* Check clipboard permissions
* Try manual copy if needed
#### "Return to Chat not working."
**Solutions:**
* Ensure chat still exists
* Check that the message ID is valid
* Try navigating manually to chat
* Chat may have been deleted
* Sometimes you have to scroll a little to find it
***
**Next:** Learn about other [interface features](/getting-started/interface-overview) in the Interface Overview.
[^1]: tag
# Usage Panel
The Usage Panel tracks your Fal.ai spending, shows usage statistics, and provides quick access to account management—all without leaving Premiere Pro.
### Where to Find It
Access the Usage Panel via:
1. **Gear icon** (⚙️) in the chat header
2. **Select "Usage"** from the menu
3. **Panel opens** showing your usage statistics
### What It Shows
#### Usage Statistics
The panel displays your Fal.ai usage across different time periods:
**Last 24 Hours:**
* Total spending in the past day
* Number of generations
* Breakdown by operation type
**Last 7 Days:**
* Weekly spending total
* Average daily usage
* Trend information
**Last 30 Days:**
* Monthly spending total
* Usage patterns
* Cost per generation averages
### Quick Actions
#### View in Fal Dashboard
**Button:** "View in Fal Dashboard" or "See Balance"
**What it does:**
* Opens your Fal.ai account dashboard in the browser
* Shows a detailed usage breakdown
* Access to billing and account settings
* View individual generation costs
#### Add Credits
**Button:** "Add Credits" or "Add Funds"
**What it does:**
* Opens Fal.ai billing page
* Add funds to your account
* Set up auto-recharge (if available)
* View payment history
#### Refresh Usage
**Button:** Refresh icon or "Refresh" button
**What it does:**
* Updates usage statistics from Fal.ai API
* Fetches the latest spending data
* Shows cached data if offline
* Syncs with your Fal.ai account
### Understanding Your Usage
#### Cost Tracking
**Why it matters:**
* **Budget management** - Know how much you're spending
* **Project costing** - Track costs per project
* **Optimization** - Identify expensive operations
* **Planning** - Estimate future costs
#### Usage Patterns
**What to look for:**
* **Peak usage times** - When you generate most content
* **Cost per generation** - Average cost per video/image
* **Model costs** - Which models are most expensive
* **Trends** - Spending increasing or decreasing
#### Cost Optimization
**Tips to reduce costs:**
* **Use faster models** - Veo 3.1 Fast vs. Sora 2 Pro
* **Turn audio off** - Disable audio generation if available
* **Lower resolution** - 720p vs. 4K when possible
* **Shorter durations** - 4s vs. 12s for videos
* **Batch operations** - Group similar requests
* **Preview before processing** - Use "Track Frame" in SAM 3
### Pay-As-You-Go Model
#### How It Works
* **No subscription** - Only pay for what you use
* **No monthly fees** - No recurring charges
* **No credit expiration** - Funds never expire
* **Transparent pricing** - See exactly what each operation costs
#### Benefits
* **Cost control** - Set your own budget
* **Flexibility** - Use as much or as little as needed
* **Project tracking** - Know costs per project
* **No commitments** - Cancel anytime
### Managing Your Account
#### Adding Funds
1. **Click "Add Credits"** in the Usage Panel
2. **Or go to:** [fal.ai/dashboard/usage-billing](https://fal.ai/dashboard/usage-billing)
3. **Select amount** - Recommended: $10-$20 to start
4. **Complete payment** - Secure payment processing
5. **Funds available immediately**
#### Recommended Starting Amount
* **$10-$20** - Good for testing and learning
* **$50** - For regular use
* **$100+** - For heavy usage or professional work
#### Auto-Recharge (If Available)
* **Set up automatic** fund addition when the balance is low
* **Never run out** of credits mid-project
* **Configure threshold** - Add funds when balance drops below X
* **Manage in the Fal.ai dashboard**
### Troubleshooting
#### "Usage not updating."
**Solutions:**
* Clickthe "Refresh" button to sync with Fal.ai
* Check the internet connection
* Verify the Fal.ai API key is correct
* Wait a few minutes (usage may be delayed)
#### "Can't add funds."
**Solutions:**
* Go tothe Fal.ai dashboard directly
* Check the payment method is valid
* Verify account permissions
* Contact Fal.ai support for billing issues
#### "Usage seems incorrect."
**Solutions:**
* Check the Fal.ai dashboard fora detailed breakdown
* Verify individual generation costs
* Some operations may be pending
* Contact Fal.ai support to review charges
### Privacy & Security
* **Usage data** is fetched from the Fal.ai API
* **No sensitive data** stored locally
* **Secure connection** to Fal.ai servers
* **Your data** is private and secure
### Related Features
* [**API Pricing**](/getting-started/pricing) - Billing through Fal.ai
* [**Settings**](/getting-started/interface-overview) - Configure API keys
* [**Generate Media** ](/getting-started/interface-overview/generate-media-button)- Where generations happen
* [**Fal.ai Dashboard**](/getting-started/interface-overview/usage-panel) - Detailed account management
***
**Next:** Try the First 5 Things to Try to start using Chat Video Pro!
# Story Cutter Assistant
Story Cutter finds the best soundbites in your footage and places them directly into your Premiere Pro timeline — saving hours of manual scrubbing.
{% embed url="" %}
#### Story Cutter analyzes your transcript, identifies the best moments based on your goals, and assembles a rough cut directly in your Premiere Pro timeline. Works with interviews, podcasts, tutorials, vlogs, event recaps, and any footage with dialogue.
{% hint style="warning" %}
**Dialogue-only tool.** Story Cutter works by reading spoken word from a transcript. It selects and arranges soundbites — it does not analyze, identify, or insert b-roll, graphics, or visual-only footage. After your rough cut is in the timeline, visual coverage is added manually in Premiere Pro as a separate step.
{% endhint %}
***
### Getting Started
#### Step 1: Prepare Your Source Timeline
Before you export anything, set up your footage in Premiere Pro the right way. This is the foundation everything else depends on.
* Put all your footage onto a **dedicated Premiere Pro timeline** — this is your source timeline
* If you're working on multiple videos, create a separate timeline for each
* If you're using multiple camera angles, **sync your clips before you transcribe** — Story Cutter can pull from **stacked synced video tracks** when you insert (especially **Insert Rough Cut**). See **Multicam and multi-angle footage** below.
* **Do not move clips on this timeline after transcribing.** Timestamps are locked to clip positions. Moving clips breaks the sync and requires a new transcription
#### Multicam and multi-angle footage
**Does Story Cutter work with multicam?**\
Yes — if by “multicam” you mean **several cameras synced on one source timeline**, each camera on **its own video track**. Story Cutter reads **regular timeline clips** on that sequence; it does **not** drive Premiere’s **multicam angle editor** inside a single multicam clip.
**How to set up your source timeline**
1. **Sync every angle first** (timecode, audio waveform, Merge Clips, etc.), **then** transcribe. If clips drift or you move them after export, timestamps won’t match — you’ll need a new transcript.
2. **Stack cameras on separate video tracks from the bottom up:** put your primary angle on **V1**, the next on **V2**, then V3 if needed.\
**Don’t leave empty video tracks below your cameras** (for example, don’t leave V1 empty while cameras only live on V3/V4). Empty gaps in the stack make it easy for an angle to be **missed** when inserts run.
3. **Confirm the linked source sequence** (purple pill) is exactly the timeline you transcribed.
**Rough cut vs. one-line insert**
* **Insert Rough Cut** walks **all** video tracks at each moment and tries to bring **multiple angles** into your edit when they’re **different clips** stacked at that timecode.
* **Inserting a single soundbite** (the **↓** on one line) picks video starting from **V1 upward** — so put the angle you want as the default for those inserts on **V1**.
**If you only have one Multicam clip on one track**
Premiere’s **multicam source sequence** often appears as **one clip** on one track. Story Cutter will treat it like **one** piece of media — it won’t switch angles inside that multicam clip. For **true multi-angle inserts**, use **stacked synced clips** on V1/V2/… as above, or stay on **one** angle for that project.
**If it “doesn’t look right”**
* Check the **purple pill** → correct source sequence.
* Check you **didn’t move** clips after transcribing.
* Try **V1 + V2 (and up) with no empty tracks under your cameras**, then **Insert Rough Cut** to verify multi-angle output.
{% hint style="info" %}
**Pro tip:** Name your source timeline clearly (e.g. "Interview - Story Cutter"). Story Cutter uses this name to auto-link your transcript when you upload it.
{% endhint %}
***
#### Step 2: Export Your Transcript
1. In Premiere Pro, open the **Text** panel (Window → Text)
2. If your footage hasn't been transcribed yet, click **Transcribe**
3. Once transcribed, click the **three dots (⋯)** in the top-right of the Text panel
4. Select **Export as transcript file**
5. Save as a `.json` file — Premiere's JSON export carries word/clip-level timestamp data, so it gives Story Cutter the most precise cuts. Plain text (`.txt`) and CSV (`.csv`) exports also work, but they carry only sentence-level timing so cuts are less precise. Recommend saving in your project's Documents folder to stay organized
{% hint style="info" %}
**Supported formats:** For the tightest cuts, use **Story Scribe** transcription (word-level timecodes) or a **JSON** transcript — Premiere Pro's native `.json` export, ElevenLabs JSON, or generic words JSON. Story Cutter also accepts **SRT**, **VTT**, and Premiere Pro plain text (`.txt`) / CSV (`.csv`) exports, though plain text and CSV carry only sentence-level timing so cuts are less precise. If your footage was transcribed elsewhere, you can import its SRT, VTT, or JSON directly — or re-transcribe with Story Scribe inside Chat Video Pro for word-level precision.
{% endhint %}
***
#### Step 3: Start Story Cutter
1. Open Chat Video Pro (Window → Extensions → Chat Video Pro)
2. On the home screen, click the **Story Cutter** conversation starter
3. Attach your transcript file using the **paperclip (📎)** button in the composer
Once attached, you'll see a small **purple pill icon** appear in the composer. This confirms your transcript is loaded and Story Cutter is ready.
***
#### Creative brief (optional)
Story Cutter works best when your **chat prompt** explains runtime, platform, and goals — but you can also supply a **creative brief** so the cut follows your plan, client notes, or show format.
* **Attach an outside document** — Use the **paperclip (📎)** to attach any reference file **in the same conversation** as your transcript. The AI reads it alongside your footage and uses it to drive selection and structure. Anything in the document — section goals, mandatory topics, brand language, pacing notes, client feedback, or show formats — feeds directly into how soundbites are chosen and ordered.
Useful documents to attach:
* Client briefs or creative direction PDFs
* Show bibles or episode formats
* Scripts or outlines from a director or producer
* Notes or feedback from a previous cut
* Previous episode structures you want to match
* **Build a brief in chat with `/creative brief`** — Type **`/creative brief`** in the composer to start a **planning-first** flow. The assistant outlines concepts, platforms, and notable moments with timecodes — **without** producing a paper cut yet. When you are ready, ask for the cut in the same thread; your brief context and prompt work together for the full pipeline.
{% hint style="info" %}
**When to use it:** Long transcripts, multi-section deliverables, or any time a plain prompt isn’t enough to capture editorial intent — for example, “Act 1 must establish the problem before we cut to the product demo.” The more context the AI has, the more intentional the selection.
{% endhint %}
***
#### Step 4: Link Your Source Timeline
When a transcript is attached, Story Cutter needs to know which Premiere Pro timeline your footage lives on. This is what allows it to insert clips in the right place.
* **Auto-link:** If your transcript file name matches your timeline name exactly, it links automatically
* **Manual link:** If they don't match, hover over the purple pill icon and click it to open the source selector — pick the correct timeline from the list
You can update the linked timeline at any point during the session by clicking the purple pill icon again.
{% hint style="info" %}
**Important:** Make sure you link the source sequence you transcribed earlier, and do not move any footage in that source sequence so it stays in sync.
{% endhint %}
***
#### Step 5: Write Your Prompt
Give Story Cutter the context it needs to make good decisions. The more specific your prompt, the better the results.
**Always include:**
| What | Why |
| ----------------------- | ------------------------------------------------------------------------ |
| **Ideal runtime** | Sets the target length (e.g. "5 minutes", "60 seconds") |
| **Platform** | Determines story structure and pacing (YouTube, TikTok, Instagram, etc.) |
| **Video type** | Vlog, tutorial, interview, event recap, documentary |
| **Hook / CTA guidance** | Tell it how you want the video to open and close |
**Optional but helpful:**
* Topics or themes to focus on
* Tone (educational, energetic, emotional, professional)
* Anything to avoid or exclude
* A **creative brief** you attached or built with **`/creative brief`** — reinforces structure and priorities alongside your message
**Example prompt:**
> "The goal of this video is to showcase Chat Video Pro and how it can help real estate editors speed up their post-production. Ideal runtime is 5 minutes, optimized for YouTube, and it's an educational tutorial. Start with an engaging hook and end with a call to action."
> **Fastest way to prompt:** Use the voice dictation button in the composer. Just talk through the goal of the video naturally — runtime, platform, type, focus. You don't need to be precise; the AI understands conversational descriptions.
***
### Working With Your Results
#### Reviewing the Paper Cut
After sending your prompt, Story Cutter scans the transcript and streams back a structured paper cut. You'll see a **thinking card** at the top — click it to watch what the AI is doing (scanning, mapping, arranging).
**Rough timing guide:**
| Transcript length | Expected time |
| ----------------- | ---------------------------------------------------------------------------------------- |
| Under 30 minutes | 10–30 seconds |
| 30–60 minutes | 30–90 seconds |
| 60–120 minutes | 1–3 minutes |
| Over 2 hours | Use Gemini 3.1 Pro — its 1M context window handles the full transcript without splitting |
The paper cut includes:
* **Soundbites** — verbatim quotes from your transcript, never paraphrased
* **Timestamps** — in `HH:MM:SS:FF` format, matching your source timeline
* **Section labels** — the AI breaks your video into named segments (Intro, Main Content, CTA, etc.)
* **Visual break markers** — placeholders for montage moments or title cards between sections
* **Editor notes** — explains why each soundbite was chosen
***
#### Navigating Soundbites
Each soundbite in the paper cut has two actions:
* **Click the timestamp** — jumps your Premiere Pro playhead to that moment on the source timeline so you can preview it
* **Click the down arrow ↓** — inserts just that one soundbite at your playhead on the active timeline
Use individual insertion when you want to hand-pick specific moments and build the edit yourself.
***
#### Insert Rough Cut
Scroll to the bottom of the paper cut and click **Insert Rough Cut** to place the entire selection onto your timeline at once.
What happens:
* All soundbites are trimmed in-point to out-point and placed in story order
* **Section markers** appear above the clips on the timeline, labeled with the AI's recommended structure
* If your footage was multi-angle and synced, Story Cutter **tries to bring stacked angles through** when each track has its own clip at that time
* Clips land starting at your playhead position
After inserting, keep the conversation going in the same thread. Story Cutter holds the full context of your transcript and previous selections.
**Refinement prompt examples:**
* "Cut 30 seconds from section 2 — it runs too long"
* "The opening doesn't land — find a stronger hook"
* "Replace the third soundbite, it feels repetitive"
* "Find a cleaner close — something that lands harder"
* "Focus section 1 more on the problem before introducing the solution"
* "Remove anything about pricing"
{% hint style="info" %}
**Before inserting:** Make sure your playhead is parked where you want the cut to begin — or use an empty sequence to keep things clean.
{% endhint %}
***
### Slash Commands
Type `/` in the chat input to open the command menu. These commands let you run targeted workflows without writing a full prompt.
***
#### `/new video`
Clears the current transcript and resets Story Cutter so you can start fresh with a different project. Use this when switching between videos in the same session.
***
#### `/social clip`
Cuts a 60-second version of your video optimized for social media — Reels, Shorts, TikTok. Story Cutter finds the best hook, builds to a clear value moment, and closes with a call to action.
**Usage:**
```
/social clip
```
Finds the single best 60-second clip from the entire transcript.
```
/social clip the rooftop interview moment
```
Focuses the social clip on a specific topic or moment.
***
#### `/top 5 soundbites`
Finds the five strongest moments in the entire transcript. Each result includes a description of why it works and which platforms it's best suited for (TikTok, Instagram Stories, LinkedIn, YouTube Shorts, etc.).
Great for pulling extra value from a shoot. Send these to your client alongside the main deliverable — it takes seconds and adds real value.
**Usage:**
```
/top 5 soundbites
```
Cost is typically a few cents. The results come back fast and each soundbite can be individually inserted with the down arrow.
***
#### `/select pass`
The most powerful command. Scans the entire transcript and groups all the best moments by category — hooks, value moments, call to actions, emotional peaks, and more. Results are labeled with section markers and can be inserted all at once or one at a time.
**Usage:**
```
/select pass
```
Finds everything and groups it all by category.
```
/select pass hooks
```
Returns only hook moments from the full transcript.
```
/select pass moments about the sky replacement demo
```
Targeted search — finds every relevant moment on a specific topic.
**Why this is powerful:** Rather than watching hours of footage to find your selects, run a select pass and let Story Cutter surface everything worth using. You make the creative decisions — the AI does the search.
{% hint style="info" %}
**Gemini 3.1 Pro is especially good for select passes** on long transcripts thanks to its large context window. Try it when working with over an hour of footage.
{% endhint %}
***
#### `/batch`
Give Story Cutter a list of edit tasks and it handles them all in one run. Instead of prompting separately for each deliverable, describe everything you need in a single message and Story Cutter works through them sequentially.
**Usage:**
```
/batch
- Cut a 5-minute YouTube edit focused on the product demo
- Pull a 60-second Instagram Reel from the Q&A section
- Find the 3 best hooks from the full interview
```
Each task runs against the same transcript. Results appear one after another in the chat — each with its own paper cut, section markers, and insert buttons.
**When to use it:**
* You have one shoot and need multiple deliverables (long-form + social clips)
* A client wants several cuts from the same footage at different lengths
* You want to run a select pass and a rough cut in one go without re-prompting
**Tips:**
* Be specific about runtime, platform, and focus for each task — the same rules that apply to a single prompt apply to each line in a batch
* Tasks run in order, so put the longest or most important edit first
* Each result is independent — inserting one doesn't affect the others
***
### Pro Tips
**Use the recommended workflow: Source → Selects → Rough Cut**
Keep your source timeline preserved and untouched. Run a `/select pass` to a dedicated selects timeline where everything is laid out with markers. Then build your rough cut from there. This gives you maximum creative control while automating the time-consuming selection work.
**Sync your footage before you transcribe**
If you're working with multi-angle footage (A-roll + B-roll, screen recording + talking head), sync everything on the source timeline before you export the transcript. When Story Cutter inserts, it brings all synced tracks — so your rough cut arrives with every angle already in place.
**Don't move clips on the source timeline**
Once transcribed, timestamps are tied to clip positions. If you rearrange clips, the timestamps break and you'll need to re-transcribe.
**Try different models for different jobs**
* **Claude Sonnet 5** — default for Story Cutter and everyday chat
* **GPT-5.5**, **Gemini 3.1 Pro**
* **Claude Opus 4.8** / **Fable 5** — chat now; Story Cutter coming soon
**Use `/batch` for multi-deliverable shoots** — One transcript, multiple outputs. Give Story Cutter a list of tasks (long-form edit, social clip, select pass) and it works through them all without re-prompting. Saves time when you need several cuts from the same footage.
**Quality in, quality out**
The more organized and specific your prompt, the better the cut. Describe the goal of the video as if you were briefing a human editor: what it's about, who it's for, what you want them to feel, and how long it should run.
***
### Troubleshooting
**Timestamps don't match the source footage** Make sure clips haven't been moved on the source timeline since transcribing. Re-transcribe if anything has shifted.
**Insert button places clips in the wrong timeline** The source timeline may not be linked correctly. Click the purple pill icon and manually select the right timeline.
**Auto-link didn't work** Happens when the transcript file name and timeline name don't match exactly. Click the purple pill to set the link manually.
**No good soundbites found** Check that your transcript has full timestamps and complete dialogue — summaries or partial transcripts won't work well. Try increasing the target duration.
**Select pass returned too many results** Add a filter to narrow it down: `/select pass hooks` or `/select pass moments about [specific topic]`.
**Multicam or multi-angle looks wrong** Confirm the **purple pill** points at the sequence you transcribed, clips weren’t moved after export, and cameras are **stacked on V1, V2, …** with **no empty video tracks below** them. A **single multicam clip** on one track counts as one layer — use stacked clips for full multi-angle rough cuts. Prefer **Insert Rough Cut** to test; single-line **↓** inserts use **V1 first**.
***
**Next:** Learn about the Color Grade Assistant for AI-powered color correction and LUT generation.
# Color Grade Assistant
The Color Grade Assistant provides professional color correction analysis and generates custom LUT files for any look you can imagine.
### Tutorial
{% embed url="" %}
### What It Does
The Color Grade Assistant:
* **Analyzes footage** using technical scope data (luminance, color cast, saturation)
* **Provides Lumetri corrections** with one-click application
* **Generates custom LUT files** (.cube format) for creative looks
* **Matches reference images** with style transfer
* **Protects skin tones** while applying creative grades
* **Emulates film stocks** (Kodak, Fuji, CineStill, etc.)
### How to Use
#### Step 1: Start the Assistant
1. Click the **Color Grading Assistant** conversation starter
#### Step 2: Export a Frame
**Method 1:** [**Frame Export Button** ](/getting-started/interface-overview/frame-capture-button)**(Recommended)**
1. In Premiere Pro, position your playhead on the clip you want to grade
2. Click the **Frame Export button** (📷) in Chat Video Pro composer
3. The frame appears as an attachment
**Method 2: Upload Screenshot**
1. Take a screenshot of your footage in Premiere Pro
2. Drag & drop or upload the image file
3. Ensure it's a high-quality capture (not compressed)
#### Step 3: Get Technical Analysis
Ask one of these:
* "What's wrong with this shot?"
* "Analyze the color."
* "How can I improve this?"
* "What corrections does this need?"
**You'll receive:**
* **Technical score** (0-100) based on exposure, color balance, and saturation
* **Specific issues** identified (underexposed, color cast, crushed shadows, etc.)
* **Lumetri recommendations**
* **One-click Apply button** to apply corrections directly
* **Custom LUT file** created for your footage if needed
Example analysis:
#### Step 4: Apply Corrections
**Option A: One-Click Apply**
1. Review the recommendations
2. Click the **"Apply to current clip"** button
3. Corrections are applied directly to your Premiere Pro clip
4. Adjust intensity in the Lumetri panel if needed
#### Step 5: Request Creative Look (Optional)
Ask for a specific style:
* "Create a cinematic LUT."
* "Give me a Blade Runner look."
* "Make it look like film noir."
* "Vintage 16mm film look"
* "Teal and orange grade"
**You'll receive:**
* **Downloadable .cube LUT file**
* **Explanation** of the look created
* **Instructions** for applying in Premiere Pro
### Style Transfer (Match Reference)
Match the color grade from a reference image to your footage.
#### Steps
1. **Upload Two Images**
* **Source** (purple border): Your footage to be graded
* **Reference** (green border): The look you want to match
2. **Use the Swap Button** (if needed)
* Click the ↔ button to switch source and reference positions
3. **Request Style Transfer**
* Type: "Match the colors" or "Apply this look to my footage"
* The AI analyzes both images and generates a matching LUT
4. **Download and Apply**
* Download the `.cube` LUT file
* Apply in Premiere: Lumetri → Creative → Look → Browse
### LUT Generation
#### Available Looks
The Color Grade Assistant can create infinite LUTs for any style. Here are some ideas:
**Cinematic Styles:**
* Blade Runner (teal shadows, warm highlights)
* Film Noir (high contrast, desaturated)
* Golden Hour (warm, soft, glowing)
* Cinematic (S-curve, teal & orange)
**Film Stock Emulation:**
* Kodak Vision3 500T
* Fuji Eterna 500
* CineStill 800T
* Kodak Portra 400
* Ilford HP5 (black & white)
**Creative Looks:**
* Vintage 16mm
* Horror/Thriller
* Documentary Natural
* Commercial Vibrant
* Desaturated Muted
#### How to Request a LUT
Be specific:
* ✅ "Create a cinematic LUT with teal shadows and warm highlights."
* ✅ "Give me a Blade Runner look."
* ✅ "Make a LUT that looks like Kodak Vision3."
* ✅ "Vintage film look with heavy grain simulation."
Less effective:
* ❌ "Make it better" (too vague)
* ❌ "Fix the colors" (use technical correction instead)
### Skin Tone Protection
When grading footage with people, the Color Grade Assistant automatically enables skin tone protection.
#### Requesting Specific Levels
* **"Heavy skin protection"** (0.7-0.85) - For close-ups, portraits
* **"Subtle skin protection"** (0.3-0.5) - For wide shots with people
* **"No skin protection"** (0) - For scenes without people, full creative freedom
#### How It Works
Skin tone protection:
* **Preserves natural skin colors** while applying creative grades
* **Adjustable intensity** based on your needs
* **Automatic detection** when people are in the frame
### Technical Analysis Details
The Color Grade Assistant uses real scope data, not visual guessing:
#### What Gets Analyzed
1. **Luminance Distribution**
* Average brightness
* Shadow/highlight clipping percentages
* Tonal range assessment
2. **Color Cast Detection**
* Warm/Cool bias
* Green/Magenta tint
* Neutral assessment
3. **Saturation Levels**
* Overall saturation
* Zone-specific saturation
* Vibrancy assessment
4. **Exposure Assessment**
* Underexposed/overexposed detection
* Safe exposure range
* Recovery recommendations
#### Score Interpretation
* **90-100:** Technically excellent, minimal corrections needed
* **70-89:** Moderate corrections recommended
* **50-69:** Significant corrections needed
* **Below 50:** Major issues, extensive correction required
### Applying LUTs in Premiere Pro
#### Method 1: Lumetri Creative Tab
1. Select your clip in Premiere Pro
2. Open the **Lumetri Color** panel
3. Go to **Creative → Look**
4. Click **Browse** next to "Look."
5. Select your downloaded `.cube` file
6. Adjust **Intensity** slider (0-200%) to taste
#### Method 2: Lumetri Effect
1. Apply **Lumetri Color** effect to your clip
2. In the Effect Controls panel
3. Go to **Creative → Look**
4. Browse and select your `.cube` file
5. Adjust intensity
### Tips for Best Results
1. **Export high-quality frames** - Use [the Frame Export button](/getting-started/interface-overview/frame-capture-button) for best accuracy
2. **Convert Log to Rec.709 first** - If working with Log footage, normalize before grading
3. **Apply Lumetri effect** - Ensure clip has Lumetri Color effect applied
4. **Be specific** - "Cinematic" is better than "make it look good."
5. **Iterate** - Ask for adjustments: "Make it warmer" or "More contrast."
6. **Combine corrections** - Apply technical fixes first, then creative LUT
### Common Workflows
#### Workflow 1: Technical Correction → Creative Grade
1. Export frame → Get analysis → Apply corrections
2. Ask for creative LUT → Download → Apply on top
3. Result: Technically correct + stylistically graded
#### Workflow 2: Style Transfer
1. Upload source + reference images
2. Request style transfer
3. Download matching LUT
4. Apply to the entire sequence
#### Workflow 3: Film Stock Emulation
1. Export frame
2. Request: "Make it look like Kodak Vision3."
3. Download film stock LUT
4. Apply and adjust intensity
### Troubleshooting
#### "Apply button doesn't appear."
* Ensure you're inthe Color Grading Assistant conversation
* Check that the frame is uploaded
* Verify the Lumetri Color effect is applied to the clip in Premiere
#### "LUT looks too strong/weak."
* Adjust **Intensity** slider in Lumetri (0-200%)
* Request a lighter/heavier version: "Make it more subtle."
* Combine with basic corrections for balance
#### "Analysis seems inaccurate."
* Use the Frame Export button (not screenshots) for best results
* Ensure the frame represents the clip accurately
* Check that the clip isn't already heavily graded
***
**Next:** Learn about the [Video Prompter Assistant](/conversation-starters/video-prompter-assistant) for structured prompt creation.
# Video Prompter Assistant
The Video Prompter Assistant helps you create structured, detailed prompts optimized for video generation models like Sora, Veo, and Kling.
### How to Use the Video Prompter Assistant
{% embed url="" %}
### What It Does
The Video Prompter Assistant:
* **Guides you through prompt creation** with structured questions
* **Recommends technique cards** with proven prompt templates
* **Optimizes prompts** for specific video models
* **Ensures completeness** (camera movement, style, setting, etc.)
* **Outputs JSON-ready prompts** for advanced workflows
* **Uses exact template phrases** from selected cards for best results
### How to Use
#### Step 1: Discovery
The assistant starts by asking:
**"What kind of video would you like to create today?"**
You'll see a table with video-type options:
**Simply describe your video idea** - for example:
* "I want a cinematic shot of a coffee shop."
* "A product ad for a new smartphone"
* "An animated logo reveal"
Attach an image to improve the accuracy of the scene description.
#### Step 2: Refinement & Recommended Cards
This is where the magic happens. After you describe your video idea, the assistant:
1. **Searches its knowledge base** for relevant video generation techniques
2. **Shows 2-4 recommended technique cards** based on your description
3. **Asks targeted refinement questions** in a table format
**Understanding Technique Cards**
**What are technique cards?**
* **Selectable prompt templates** with proven phrasing that AI video models understand
* **Pre-written templates** with placeholders like `[Subject]`, `[Action]`, `[Setting]`
* **From knowledge base** - Curated techniques from professional video generation workflows
* **Custom cards** - Sometimes created specifically for your unique request
**Card types:**
* **Camera cards** - Camera movements and techniques (push-in, dolly, orbit, etc.)
* **Effect cards** - Visual effects and techniques (slow motion, time-lapse, etc.)
* **Template cards** - Complete shot templates with all elements
**Example card:**
```
Card Name: "Slow Push-In"
Description: "Cinematic approach shot."
Template: "Slow push-in on [Subject] performing [Action] in [Setting],
capturing [Key Moment] with [Visual Detail]."
```
**How to Use Cards**
1. **Review the recommended cards** - They appear as selectable boxes in the chat
2. **Click cards to select them** - Selected cards highlight (you can select multiple)
3. **Cards contain templates** - Each card has a template with placeholders
4. **Answer refinement questions** - The assistant asks 2-3 targeted questions in a table
5. **Provide details** - Your answers fill in the template placeholders
**Example refinement table:**
| **Question** | **What I Need** |
| -------------------- | --------------------------------------------- |
| **Subject Details:** | What unique traits should the character have? |
| **Setting:** | Where does this take place? |
| **Camera Movement:** | How should the camera move? |
**Card Selection Strategy**
**Select cards when:**
* ✅ They match your vision closely
* ✅ You want proven techniques that work well
* ✅ The template phrasing sounds right for your video
* ✅ You want to combine multiple techniques
**Skip cards when:**
* ❌ None match your specific needs
* ❌ You have a very unique vision
* ❌ You prefer to describe everything yourself
**Note:** You can proceed without selecting cards - the assistant will create a custom prompt based on your description.
#### Step 3: Synthesis
**When synthesis happens:**
* After you've selected technique cards (if any) AND
* You indicate readiness (e.g., "continue", "ready", "let's do it", "create the prompt")
**What happens:**
1. **Template integration** - If you selected cards, their template text is used EXACTLY (verbatim)
2. **Placeholder filling** - Your refinement answers fill in the `[Subject]`, `[Action]`, etc.
3. **JSON generation** - Everything is combined into a polished JSON prompt under 1000 characters
4. **Model optimization** - The prompt is optimized for the recommended video model
**Output format:** You'll receive a complete JSON prompt ready to use:
```json
{
"description": "[Detailed description incorporating selected techniques, visual style, action, environment]",
"camera": {
"motion": "[Camera motion - uses exact template phrases from selected cards]",
"angle": "[Camera angle]",
"composition": "[Composition]"
},
"audio": {
"dialogue": "[Dialogue or null]",
"sfx": "[Sound effects]",
"music": "[Music style]"
},
"negative_prompt": "[Things to avoid - blur, distortion, etc.]",
"character_details": "[Character description if .applicable, otherwise null]"
}
```
#### Step 4: Use the Prompt
1. **Copy the JSON prompt** from the code block
2. **Open Generate Media** in Chat Video Pro
3. **Select the recommended model** (Sora, Veo, Kling, etc.)
4. **Paste the full JSON** into the prompt field — Chat Video Pro **automatically extracts the `description` field** for generation, so you don't need to manually pull out just the description text
5. **Generate your video**
**Optional:** If you want to refine the prompt before generating, copy just the `"description"` string and edit it directly.
#### How Templates Work
**Template example:**
```
"Slow push-in on [Subject] performing [Action] in [Setting],
capturing [Key Moment] with [Visual Detail]."
```
**After filling placeholders:**
```
"Slow push-in on a barista preparing espresso in a cozy coffee shop,
capturing the steam rising from the cup with a shallow depth of field."
```
**Why this works:**
* "Slow push-in" is a specific camera language that AI models recognize
* The structure ensures all elements are included
* The phrasing is optimized for video generation APIs
### Complete Workflow Example
#### Example 1: With Card Selection
1. **User:** "I want a cinematic coffee shop video."
2. **Assistant shows cards:**
* "Slow Push-In" (selected)
* "Golden Hour Lighting" (selected)
* "Shallow Depth of Field" (not selected)
3. **Assistant asks:**
* What's the main subject? → "A barista making espresso."
* What's the setting? → "Cozy urban coffee shop.p"
* What time of day? → "Early morning, golden hour.r"
4. **Synthesis creates JSON** using the exact template phrases from selected cards
5. **User generates a video** with an optimized prompt
#### Example 2: Without Card Selection
1. **User:** "I want a unique video of a robot dancing in space."
2. **Assistant shows cards:**
* "Orbit Camera" (related, not selected)
* "Slow Motion" (related, not selected)
3. **User:** "Just create it based on my description."
4. **Assistant creates a custom prompt** without using card templates
5. **User generates a video** with a custom-tailored prompt
### Prompt Structure
A well-structured video prompt includes:
#### Essential Elements
1. **Subject** - Clear description of main focus
2. **Setting** - Location, time, environment
3. **Action** - What's happening, movement
4. **Camera** - Movement, angle, framing (often from cards)
5. **Style** - Visual aesthetic, mood
6. **Details** - Lighting, colors, atmosphere
### Model-Specific Optimization
The assistant tailors prompts for different models:
#### Sora 2
* Emphasizes cinematic language
* Works well with detailed scene descriptions
* Supports longer prompts (up to 12 seconds)
* Cards often include "cinematic" and "film grain" templates
#### Veo 3.1
* Focuses on dialogue and audio cues
* Optimizes for character consistency
* Great for narrative scenes
* Cards emphasize character and dialogue templates
#### Kling 3.0 / O3
* Emphasizes camera movement and cinematic motion
* Works well with dynamic action and human subjects
* Supports complex camera moves (360°, FPV, orbit, dolly)
* Kling 3.0 Pro/Standard for universal use (T2V, I2V, transition)
* Kling O3 for advanced motion understanding and character reference
* Cards include advanced camera technique templates
#### Hailuo 03
* Optimizes for action and motion
* Great for sports and dynamic scenes
* Emphasizes fluid movement
* Prompts the soundscape too, since Hailuo 03 always returns native stereo audio
* Cards focus on motion and action templates
### Advanced: JSON Prompt Format
For advanced users, the assistant outputs JSON-ready prompts:
```json
{
"description": "Slow push-in on a barista preparing espresso...",
"model": "sora-2",
"aspect_ratio": "16:9",
"duration": 8,
"style": "cinematic",
"camera_movement": "push-in",
"lighting": "golden hour"
}
```
**Key fields:**
* `description` - Full prompt text (incorporates card templates)
* `camera.motion` - Exact camera movement from selected cards
* `audio` - Dialogue, SFX, or music cues
* `negative_prompt` - Things to avoid
* `character_details` - Character descriptions, if applicable
### Tips for Best Results
1. **Select relevant cards** - Choose cards that match your vision
2. **Be specific in answers** - "Cozy coffee shop" is better than "coffee shop."
3. **Combine multiple cards** - Select 2-3 cards for richer prompts
4. **Trust the templates** - Card templates use proven phrasing
5. **Answer all questions** - More details = better prompts
6. **Review the JSON** - Check that selected card templates are included
7. **Iterate if needed** - Ask for adjustments or try different cards
### Common Use Cases
* **Product showcases** - Detailed product videos with cinematic cards
* **B-roll creation** - Background footage using camera movement cards
* **Scene establishment** - Setting the mood with lighting/style cards
* **Character introductions** - Character-focused shots with portrait cards
* **Transition elements** - Between-scene footage using transition cards
### Troubleshooting
#### "No cards are showing."
**Solutions:**
* The assistant may not have found relevant techniques
* Try being more specific about your video type
* The assistant will create a custom prompt without cards
* You can still get excellent results without cards
#### "Cards don't match what I want."
**Solutions:**
* Don't select cards that don't fit
* Answer refinement questions with your specific vision
* The assistant will create a custom prompt
* You can ask for different card suggestions
#### "Prompt is too vague."
**Solutions:**
* Select technique cards for structure
* Answer all refinement questions thoroughly
* Add more specific details in your description
* Ask the assistant to expand on specific elements
#### "Selected cards aren't in the final prompt."
**Solutions:**
* Check that you actually selected the cards (they should highlight)
* Make sure you said "continue" or "ready" after selecting
* The templates should appear verbatim in the JSON
* If missing, ask the assistant to include the selected techniques
#### "Model recommendation doesn't match my needs."
**Solutions:**
* Specify your requirements clearly in Step 1
* Mention if you need audio (Veo is best)
* State duration preferences
* Describe camera movement needs (Kling for complex moves)
***
**Next:** Learn about the[ Brand Voice Assistant](/conversation-starters/brand-voice-assistant) for consistent brand content.
# Brand Voice Assistant
### How to Use the Brand Voice Assistant
{% embed url="" %}
### What It Does
The Brand Voice Assistant:
* **Builds your brand profile** through a simple 3-step wizard
* **Generates platform-specific copy** (YouTube titles, Instagram captions, etc.)
* **Maintains consistent tone** across all content
* **Suggests music and captions** that match your brand
* **Creates SEO-optimized content** with keywords and hashtags
### How to Use
#### Step 1: Start the Assistant
1. Click the **Brand Voice Assistant** conversation starter
2. The setup wizard appears automatically
#### Step 2: Complete the 3-Step Wizard
#### Step 3: Save Your Profile
Click **"Complete Setup"** to save your brand profile. The chat is automatically renamed to "\[Brand Name] brand voice" for easy identification.
#### Step 4: Generate Content
Once your profile is set up, you can:
**Generate YouTube Content**
**Commands:**
* `/copy` or "Create YouTube copy for this transcript"
* Upload a transcript or describe your video
**Output:**
* **5 title options** (optimized for SEO and clicks)
* **Full description** (with keywords and structure)
* **Chapters with timestamps** (if transcript provided)
* **Comma-separated tags** (for YouTube tags field)
**Generate Instagram/TikTok Content**
**Commands:**
* `/copy` or "Create Instagram caption"
* Upload transcript or describe content
**Output:**
* **Long caption** (full version with context)
* **Short caption** (for posts with character limits)
* **15 hashtags** (relevant and on-brand)
* **Hook options** (attention-grabbing opening lines)
**Generate SEO Content**
**Commands:**
* `/seo` or "Create SEO content for this"
**Output:**
* **Primary keywords**
* **Secondary keywords**
* **Hashtag strategy**
* **Meta descriptions**
* **Content optimization tips**
**Find Music**
**Commands:**
* `/music` or "Find music for this video"
**Output:**
* **Music recommendations** based on brand voice
* **Vibe suggestions** (energetic, calm, cinematic, etc.)
* **Platform-specific music** (royalty-free options)
### Platform-Specific Outputs
YouTube
**Titles:**
* SEO-optimized
* Click-worthy
* Brand-consistent tone
* Multiple variations
**Description:**
* Hook paragraph
* Video summary
* Key points/timestamps
* CTA aligned with brand goal
* Links and resources
**Chapters:**
* Auto-generated from transcript
* Timestamp format: `00:00:00 - Chapter Name`
* Organized by content flow
**Tags:**
* Comma-separated
* Mix of broad and specific
* Brand-relevant keywords
Instagram
**Long Caption:**
* Full story with context
* Brand voice throughout
* Natural hashtag integration
* CTA at the end
**Short Caption:**
* Condensed version
* Essential info only
* Perfect for Reels/Stories
**Hashtags:**
* 15 relevant tags
* Mix of popular and niche
* Brand-aligned
TikTok
**Caption:**
* Hook-focused
* Engaging and on-trend
* Brand voice adapted for the platform
* CTA integrated naturally
**Hashtags:**
* Trending + niche mix
* Platform-appropriate
### Editing Your Profile
To update your brand voice profile:
1. Type `/setup` in the chat
2. The wizard reappears with current values pre-filled
3. Make changes and save
### Skipping Setup
If you skip the wizard:
* Type `/setup` anytime to configure
* Or just start using commands - the assistant will prompt for setup if needed.
### Advanced Features
#### Website Analysis
The Brand Voice Assistant can analyze your website to extract:
* Brand voice automatically
* Tone and style
* Key messaging
* Content patterns
**How to use:**
* Open the /setup and input your website
* The assistant extracts brand information automatically
#### Music Recommendations
Based on your brand profile, get music suggestions:
* **Vibe matching** - Music that fits your brand tone
* **Platform optimization** - Music suitable for your target platforms
* **Trend research** - Current popular tracks in your niche
#### Content Variations
Request multiple versions:
* "Give me 3 different Instagram captions"
* "Create 5 YouTube title options"
* "Show me hook variations"
### Tips for Best Results
1. **Complete the wizard fully** - More info = better output
2. **Be specific in brand notes** - Include phrases to use/avoid
3. **Upload transcripts** - Best results come from actual content
4. **Iterate** - Ask for adjustments: "Make it more casual" or "Add more energy"
5. **Use platform-specific commands** - `/copy` for general, `/seo` for SEO-focused
### Common Workflows
#### Workflow 1: YouTube Video Launch
1. Complete brand voice setup
2. Upload video transcript (export your `.json` transcript from the Premiere Pro Text panel, or paste the content directly)
3. Run `/copy` command
4. Get titles, description, chapters, tags
5. Copy to YouTube upload page
{% hint style="success" %}
**Just finished Story Cutter?** Your `.json` transcript file is a perfect input here. Drag it into the composer and run `/copy` — the assistant reads the transcript and generates all your metadata in your brand voice.
{% endhint %}
#### Workflow 2: Instagram Post
1. Describe your content or upload a transcript
2. Request an Instagram caption
3. Get long + short versions + hashtags
4. Use the appropriate version for your post type
#### Workflow 3: Multi-Platform Campaign
1. Set up brand voice
2. Create core content description
3. Generate platform-specific versions:
* YouTube: `/copy`
* Instagram: "Create an Instagram version."
* TikTok: "Create TikTok version"
### Troubleshooting
#### "Setup wizard doesn't appear."
* Click the Brand Voice Assistant starter again
* Type `/setup` to manually trigger
* Check that you're in a new chat (wizard only shows once)
#### "Content doesn't match my brand."
* Review your brand notes section
* Be more specific about tone and style
* Use `/setup` to refine your profile
* Provide examples: "Write like this: \[example]"
#### "Platform output format is wrong."
* Verify you selected the correct platforms in setup
* Use explicit commands: `/copy` for YouTube, "Instagram caption" for Instagram
* The assistant adapts based on your request
***
**Next:** Explore [Video Generation Features](/features/video-generation) to create AI-powered videos.
# Premiere Pro Guru
The Premiere Pro Guru is your expert assistant for troubleshooting, learning, and optimizing Adobe Premiere Pro.
### How to Use the Premiere Pro Guru
{% embed url="" %}
### What It Does
The Premiere Pro Guru:
* **Fixes technical issues** - Troubleshoot crashes, bugs, errors, and timeline problems
* **Teaches workflows** - Learn Lumetri, captions, multicam, proxies, audio, and more
* **Optimizes performance** - Speed up playback, exports, and project loading
* **Finds tutorials** - Locate high-quality online guides and resources using web search
### How to Use
#### Step 1: Start the Assistant
1. Click the **Premiere Pro Guru** conversation starter
2. Or start a new chat and ask any Premiere Pro question
#### Step 2: Choose Your Help Category
The Guru will ask you to select from four categories:
Option
Category
What It Helps With
1
Troubleshooting
Fix crashes, bugs, errors, timeline issues
2
Learn a Tool
Teach me a feature or workflow
3
Speed & Performance
Optimize Premiere Pro and hardware
4
Tutorials & Guides
Find high-quality resources online
#### Step 3: Get Help Based on Your Category
**Category 1: Troubleshooting**
**What the Guru asks:**
* "What problem are you running into? Describe what's going wrong and when it happens."
* "What version of Premiere are you using?"
* "Mac or Windows?"
* "Any third-party plugins installed?"
**What you'll get:**
* **Step-by-step diagnostic** with numbered fixes
* **Workarounds** for known bugs
* **Links to official Adobe resources** (via web search)
* **Confirmation steps** to verify the fix worked
**Example scenarios:**
* Premiere Pro crashes on export
* Timeline playback is choppy
* GPU acceleration not working
* Audio sync issues
* Project file corruption
**Category 2: Learn a Tool**
**What the Guru asks:**
* "Which tool or workflow do you want to learn? (e.g., Lumetri, captions, proxies, multicam)"
**What you'll get:**
* **Mini-lesson** with step-by-step instructions
* **Checklist** for the workflow
* **Google and YouTube search links** for more examples
* **Best practices** and tips
**Example topics:**
* Lumetri Color grading
* Caption workflows
* Multicam editing
* Proxy creation
* Audio mixing
* Essential Graphics
* Speed ramping
* Nesting sequences
**Example output:**
```
📘 Mini Lesson: Speed Ramping in Premiere Pro
1. Select your clip in the timeline
2. Right-click → Show Clip Keyframes → Time Remapping → Speed
3. Use the Pen Tool to add keyframes where you want the speed change
4. Drag speed segments up/down to change speed
5. Smooth transitions by dragging the handles between keyframes
🔍 Want more examples?
[Google Search - speed ramping Premiere Pro]
[YouTube Search - speed ramping Premiere Pro]
```
**Category 3: Speed & Performance**
**What the Guru asks:**
* "What feels slow — playback, exporting, project load time, something else?"
* "What's your system setup? (CPU, GPU, RAM, storage type)"
**What you'll get:**
* **Specific optimization steps** tailored to your issue
* **Hardware acceleration** setup guidance
* **Proxy workflow** walkthrough
* **Cache optimization** tips
* **Codec recommendations** (H.264 to ProRes transcoding)
* **Storage optimization** (moving cache to SSD)
* **Performance guides** from Adobe and pro editors (via web search)
**Common optimizations:**
* Enable hardware acceleration
* Create proxy files for 4K/8K footage
* Transcode H.264 to ProRes for smoother editing
* Move media cache to SSD
* Adjust render settings
* Optimize project structure
**Category 4: Tutorials & Guides**
**What the Guru asks:**
* "What do you want to learn or fix? Be specific so I can find the right guide."
**What you'll get:**
* **Curated search results** from web search
* **Adobe official guides** (when available)
* **YouTube tutorials** from reputable sources
* **Blog walkthroughs** from professional editors
* **Short descriptions** of what each resource covers
* **Why it's a good match** for your request
**The Guru uses Gemini web search** to find:
* Official Adobe documentation
* High-quality YouTube tutorials
* Professional blog posts and guides
* Community forum solutions
### Available Commands
You can also use direct commands for faster access:
Command
What It Does
/troubleshoot
Guided diagnostic flow to fix bugs, crashes, and errors. Uses web search for known issues & fixes.
/learn
Teach a specific tool or workflow with steps, checklists, and optional external tutorials.
/optimize
Performance tuning for smoother editing & faster exports. May include proxy setup, cache optimization, hardware tips.
/findguide
Use web search to locate the best tutorials and guides online for your exact request.
/help
Remind you of all available commands and what they do.
#### Using Commands
**Example:**
```
/learn speed ramping
```
The Guru will:
1. Provide a mini-lesson with steps
2. Give you a checklist
3. Provide Google and YouTube search links
### How the Guru Works
#### Expert Knowledge
The Premiere Pro Guru acts as an **Adobe-certified expert** with 15+ years of professional experience. It:
* Has fixed every glitch, timeline bug, GPU crash, and render hang
* Has trained thousands of users at every skill level
* Provides clear, logical, empathetic responses
* Avoids vague advice — always gives specifics
* Rephrases simply if you seem confused
#### Web Search Integration
The Guru uses **web search** (via Gemini) to:
* Find current solutions for known bugs
* Locate the best tutorials for any topic
* Access official Adobe resources
* Discover community solutions
**Important:** The Guru provides **search links** (Google and YouTube) rather than direct tutorial URLs, so you can find the most current and relevant resources.
#### Response Format
The Guru uses:
* **Beautiful markdown** formatting for easy reading
* **Emoji labels:** 📘 Learning, ⚠️ Warning, ✅ Fix, 🚀 Speed Tip, 🔍 Search Result
* **Bold numbers** for step-by-step instructions
* **Tables** for organized information
* **Concise, actionable** sentences
### Common Use Cases
#### Troubleshooting
* **Premiere Pro crashes** - Get step-by-step fixes and workarounds
* **Playback issues** - Optimize settings and hardware acceleration
* **Export problems** - Fix codec, format, or rendering errors
* **Timeline bugs** - Resolve sync, nesting, or performance issues
* **Audio problems** - Fix sync, mixing, or format issues
#### Learning Workflows
* **Color grading** - Learn Lumetri Color panel
* **Caption creation** - Master caption workflows
* **Multicam editing** - Set up and edit multicam sequences
* **Proxy workflows** - Create and use proxy files
* **Audio mixing** - Learn Essential Sound panel
* **Graphics** - Master Essential Graphics panel
* **Advanced editing** - Speed ramping, nesting, compositing
#### Performance Optimization
* **Slow playback** - Enable hardware acceleration, create proxies
* **Long export times** - Optimize render settings, use Media Encoder
* **Project loading delays** - Optimize project structure, clear cache
* **System bottlenecks** - Hardware recommendations, storage optimization
#### Finding Resources
* **Official Adobe guides** - Access current documentation
* **Video tutorials** - Find YouTube walkthroughs
* **Community solutions** - Discover forum answers
* **Best practices** - Learn from professional editors
### Tips for Best Results
1. **Be specific** - The more details you provide, the better the help
2. **Mention your version** - Premiere Pro version helps diagnose issues
3. **Describe when it happens** - Context helps identify the problem
4. **List your plugins** - Third-party plugins can cause conflicts
5. **Share system specs** - For performance issues, hardware info is crucial
6. **Use commands** - Direct commands (`/learn`, `/troubleshoot`) for faster access
7. **Follow up** - Ask for clarification if something isn't clear
### Example Interactions
#### Troubleshooting Example
**You:** "Premiere keeps crashing when I export"
**Guru asks:**
* What version of Premiere?
* Mac or Windows?
* What format are you exporting to?
* Any plugins installed?
**Guru provides:**
* Step-by-step diagnostic
* Known bug workarounds
* Links to Adobe forums
* Verification steps
#### Learning Example
**You:** "I want to learn multicam editing"
**Guru provides:**
* Overview of multicam workflow
* Step-by-step setup guide
* Checklist for editing
* Google and YouTube search links
#### Performance Example
**You:** "Playback is choppy with 4K footage"
**Guru asks:**
* System specs (CPU, GPU, RAM)
* Storage type (HDD vs SSD)
* Codec of source footage
**Guru provides:**
* Proxy creation walkthrough
* Hardware acceleration setup
* Codec transcoding recommendations
* Performance optimization guides
### What Makes It Different
Unlike generic help, the Premiere Pro Guru:
* **Stays in context** - Works inside Premiere Pro via Chat Video Pro
* **Uses web search** - Finds current solutions and tutorials
* **Provides specifics** - Never gives vague advice
* **Adapts to you** - Rephrases if you're confused
* **Offers follow-up** - Confirms fixes worked, suggests optimizations
### When to Use
Use the Premiere Pro Guru when you:
* **Hit a technical wall** - Crashes, errors, bugs
* **Want to learn** - New tools, workflows, features
* **Need speed** - Slow playback, exports, or loading
* **Want resources** - Tutorials, guides, best practices
**Don't use it for:**
* Chat Video Pro-specific questions (use regular chat)
* AI generation questions (use other assistants)
* General video editing theory (use web search directly)
***
### Related Pages
* [Story Cutter Assistant](/conversation-starters/story-cutter-assistant) - Cut stories from transcripts
* [Color Grade Assistant](/conversation-starters/color-grade-assistant) - Professional color grading
* [Video Prompter Assistant ](/conversation-starters/video-prompter-assistant)- Optimize video prompts
* [Brand Voice Assistant](/conversation-starters/brand-voice-assistant) - Maintain brand consistency
* [Interface Overview ](/getting-started/interface-overview)- Learn the interface
# Video Generation
Chat Video Pro provides powerful AI video generation capabilities, allowing you to create videos from text prompts, animate images, create smooth transitions, and use reference images for consistent c
### How to Use Generative Video to Amplify Client Videos
{% embed url="" %}
### Overview
Video Generation is where you create new video, animate images, build transitions, use reference images, extend existing clips, and open the classic video editor when you are already working from a generated or imported result.
Use this section when you want to:
* Create a video from a text prompt.
* Animate a still image.
* Generate a transition between two frames.
* Use reference images for character or style consistency.
* Extend an existing clip.
* Edit a generated or imported video from the classic editor.
For guided workflows like Cinematic Lab, Motion Director, AI Transitions, Rotoscoping, Erase Objects, Add Effects, Reshoot, Upscale, Motion Capture, Multi-Cam, and Relight Scene, use Studio.
***
### Studio vs. Video Generation
Video Generation and Studio overlap because they both create or modify video, but they are designed for different entry points.
Start here
Use when
Video Generation
You want the classic prompt/model workflow, are choosing a video model directly, or are working from an existing chat result.
Studio
You want a guided creative workflow with the right asset loader, model, presets, and controls already selected.
Examples:
Goal
Best starting point
Generate a shot from scratch with a video model
Text-to-Video
Animate a still image with a direct prompt
Image-to-Video
Continue a clip by 7 seconds
Generative Extend
Open a generated video and make a quick classic edit
Video Canvas Editor
Build a directed image-to-video shot with movement presets
Studio Motion Director
Create a polished first/last frame transition
Studio AI Transitions
Remove an object, rotoscope, reshoot, relight, or upscale
Studio Post-Production workflows
The simple rule: use **Video Generation** when you want direct model control, and use **Studio** when you want the tool to guide the workflow.
***
### Getting Started
1. Enable **Generate Media** in the composer.
2. Choose a video model from the model dropdown.
3. Choose the mode based on your inputs: text only, one image, two images, or references.
4. Configure aspect ratio, duration, resolution, and audio options.
5. Write your prompt.
6. Generate the video.
7. Use **Edit**, **Extend**, or **Studio** if the result needs another pass.
If you are starting with footage from Premiere instead of a prompt, use Studio or import the clip first, then open the classic editor from the video thumbnail.
***
### Quick Reference
***
### Workflow Tips
#### Choose The Mode From Your Inputs
If you have no image, use Text-to-Video. If you have one image, use Image-to-Video. If you have a start and end frame, use Transition Mode. If you have multiple identity or style references, use Reference Mode.
#### Use Studio For Polished Workflows
Studio is better when the task has a known creative shape: motion direction, cinematic image creation, AI transitions, rotoscoping, object removal, relighting, reshooting, VFX, multi-cam generation, motion capture, or upscaling.
#### Edit After You Generate
A strong generation often becomes better after one extra pass. Use the classic Video Canvas Editor for quick follow-up edits, or open Studio when the next step is one of the guided workflows.
#### Upscale Last
Do not upscale every draft. Generate and revise first, then use Studio Upscale when the clip is approved and needs a finishing pass.
***
### Next Steps
* Learn which model to choose in Supported Video Models.
* Start from scratch with Text-to-Video.
* Animate a frame with Image-to-Video.
* Continue a clip with Generative Extend.
* Use Studio for guided production and post-production workflows.
# Supported Video Models
Chat Video Pro supports multiple state-of-the-art AI video generation models, each optimized for different use cases.
Chat Video Pro includes several video models because no single model is best at everything. Some are better for dialogue. Some are better for cinematic camera movement. Some are better for fast drafts, longer clips, reference consistency, or mobile-first content.
This page is a practical chooser guide. You do not need to memorize every model. Start from the type of shot you want, then choose the model that fits the job.
{% hint style="info" %}
If you are using **Studio**, the workflow often chooses the model path for you. For example, Motion Director uses Kling 3.0 Pro Image-to-Video, AI Transitions uses Kling O3 transition models, and Add Effects uses Kling O3 VFX. Use this page when you are choosing models directly from the Video Generation model selector.
{% endhint %}
***
### Quick Recommendations
If you need...
Start with...
Why
Dialogue, speech, or generated audio
Veo 3.1 or Veo 3.1 Fast
Strongest choice when audio and lip-sync matter.
Fast lower-cost Veo drafts without audio
Veo 3.1 Lite
Good for visual drafts, B-roll ideas, and transitions when you will add sound in Premiere.
Cinematic motion and strong general quality
Kling 3.0 Pro or Kling O3 Pro
Strong camera movement, action, and polished motion.
Longer cinematic clips without audio
Sora 2 or Sora 2 Pro
Good for cinematic B-roll and longer visual generations.
Natural motion with ambient audio
Seedance 2
Good all-around model for natural motion and audio-enabled scenes.
Action, sports, or fast movement
Hailuo 03
Strong for dynamic action and energetic movement, at 2K with native audio.
Reference from a video or audio clip
Hailuo 03 Reference or Seedance 2.5 Reference
Both take reference video and reference audio, not just stills. Hailuo 03 caps at 12 files combined and renders 2K; Seedance 2.5 accepts up to 50 references combined and runs to 30s at up to 1080p.
Fast content with audio built in
Grok Imagine 1.5
Fast generations up to 1080p with audio always on.
High-resolution 1080p-style outputs
Wan 2.7
Good when clarity, flexible aspect ratios, or reference workflows matter.
A guided camera move from a still image
Studio Motion Director
Easier than hand-prompting image-to-video movement.
A polished transition between two frames
Studio AI Transitions
Easier than manually choosing a transition model.
Native 4K video output
Seedance 2 (4K resolution pill)
Full 3840×2160 without separate upscale.
Fast Seedance iterations
Seedance 2 Mini
Seedance-quality at lower cost.
Audio and video in one pass
Google Omni Flash
Synchronized audio.
Fast mobile-first with lipsync
Grok Imagine 1.5
1080p image-to-video with lipsync.
2K output with native audio
Hailuo 03
2K on every tier, 5–15s, stereo audio always on.
A clip of 15–20s at 1080p
Flux 3
5–20s with native audio at 720p or 1080p.
A clip that runs past 20s
Seedance 2.5
4–30s in a single pass with native audio, at 480p, 720p, or 1080p.
A reference set larger than 12 files
Seedance 2.5 Reference
Accepts up to 50 references combined across images, videos, and audio.
Ultra-wide 2:1 or 21:9 framing
Flux 3
Eight aspect ratios, and no other video model offers 2:1.
A silent cinematic plate with a cheap draft tier
Luma Ray 3.2
5s or 10s, 540p/720p/1080p, no audio pill anywhere — score it in Premiere.
Restyle a clip with control over how far it diverges
Luma Ray 3.2 Edit
Video-to-video with a 4-way divergence dial: Default, Adhere, Balanced, Reimagine.
***
### The Simple Rule
Choose the model based on the hardest part of the shot:
Hardest part of the shot
What to prioritize
A person speaking
Audio and lip-sync.
Complex camera movement
Motion quality and scene understanding.
A client-ready cinematic insert
Quality and consistency.
A quick concept
Speed and cost.
A character must look the same
Reference mode or Studio workflow.
Two frames need to connect
Transition mode or Studio AI Transitions.
The clip must be vertical/mobile
Aspect ratio support.
The best model is not always the highest-quality model. The best model is the one that solves the specific problem in the shot.
***
### Model Guide
#### Veo 3.1
Use Veo 3.1 when the shot needs audio, dialogue, or speaking characters.
Best for:
* Talking-head concepts.
* Product explainers with speech.
* Scenes where sound matters.
* Short dialogue tests.
* Image-to-video or transition shots where audio should be part of the generation.
Use **Veo 3.1 Fast** when you want quicker iterations. Use the regular Veo 3.1 path when quality matters more than speed.
#### Veo 3.1 Lite
Use Veo 3.1 Lite when you want a lower-cost Veo-style visual draft and do not need generated audio.
Best for:
* Silent B-roll concepts.
* Visual drafts before adding voiceover or music in Premiere.
* Budget-conscious text-to-video, image-to-video, or transition tests.
* Shots where 720p or 1080p is enough.
Avoid Lite when the prompt depends on spoken dialogue or synchronized audio. Add sound in Premiere instead.
#### Kling 3.0
Use Kling 3.0 when you want strong cinematic movement, flexible shot types, and polished video generation.
Best for:
* Camera moves.
* Dynamic product shots.
* Cinematic B-roll.
* Image-to-video with strong motion.
* Shots that need native audio but are less dialogue-focused than Veo.
Kling 3.0 is a strong general-purpose choice for classic text-to-video and image-to-video generation.
#### Kling O3
Use Kling O3 when motion quality, scene understanding, or transition quality is the priority.
Best for:
* Advanced motion.
* High-quality transitions.
* Reference-driven shots.
* VFX-oriented generations.
* Complex scenes where the model needs to preserve structure.
If you are creating transitions, consider Studio AI Transitions instead of manually selecting an O3 transition model. The Studio workflow gives you transition styles and better prompting structure.
#### Sora 2
Use Sora 2 when you want cinematic visual quality and do not need generated audio.
Best for:
* Establishing shots.
* Atmospheric B-roll.
* Longer visual clips.
* Cinematic concepts where dialogue is not required.
Choose Sora 2 Pro when you want the higher-quality Sora option and the extra cost makes sense.
#### Seedance 2
Use **Seedance 2** when you want natural motion with native audio and a balanced all-around video model. Select the **4K resolution pill** for full 3840×2160 output without a separate upscale pass.
Best for:
* Natural movement.
* Ambient audio scenes.
* Cinematic clips with sound.
* Reference mode when you need subject consistency.
* Wide or cinematic aspect ratios.
* Native 4K delivery when the resolution pill is enabled.
Use **Seedance 2 Mini** for faster, lower-cost iterations when you want Seedance-quality motion without the full standard price.
Use **Seedance 2 Fast** when you want the same general family with quicker turnaround.
#### Seedance 2.5
Use **Seedance 2.5** when the clip needs to run longer than the rest of the Seedance family, or when a reference set is bigger than other models will take.
Seedance 2.5 ships **alongside** Seedance 2, not as a replacement. It runs to 30 seconds at up to **1080p** (default 720p), where Seedance 2's standard tier also reaches 1080p but adds **4K** — so which one is "better" depends on whether you need running time or native 4K. Both sit in the same **Seedance** submenu.
Best for:
* Clips of **4–30 seconds** in a single pass, with native audio.
* Letting the model choose the length — **Auto** is the default duration.
* Several shots inside one generation, sequenced by describing the cuts in the prompt.
* Start-frame-to-end-frame morphs that need more than 15 seconds.
* Reference-driven work with a large or mixed reference set.
There are **two picker entries**: **Seedance 2.5** (text-to-video and image-to-video on one entry) and **Seedance 2.5 Reference**.
* **Duration:** 4–30 seconds, whole seconds, or **Auto** — and Auto is the default.
* **Resolution:** **480p, 720p, or 1080p**, defaulting to 720p. There is no 4K on 2.5.
* **Audio:** native, generated with the picture, **On** by default and genuinely toggleable — the same pill the Seedance 2 tiers use. It is ambient and scene audio, not dialogue-grade lip-sync.
* **Aspect ratios:** auto (default), 21:9, 16:9, 4:3, 1:1, 3:4, 9:16 — the same seven the Seedance 2 tiers offer.
* **Two images:** the second becomes the end frame **on the same image-to-video endpoint**. Nothing re-routes, and there is no separate transition entry.
* **References:** the composer shows 9 image, 3 video, and 3 audio slots. The model itself accepts up to **50 references combined** across all three types, so that budget is headroom at the API rather than something today's composer can fill.
{% hint style="warning" %}
There is **no Seedance 2.5 Fast and no Seedance 2.5 Mini** — 2.5 has one quality tier plus Reference. Those speed variants exist only for Seedance 2. Multi-shot sequencing is likewise **prompt-driven**: describe the cuts you want. There is no multi-shot control anywhere in the UI.
{% endhint %}
Seedance 2.5 is billed by tokens rather than a flat per-second rate: **$0.0214 per 1,000 tokens**, where tokens are `height × width × seconds × 24 ÷ 1024`. Because the frame dimensions are in the formula, the per-second cost varies with the aspect ratio you choose. At 16:9 that works out to roughly **$0.4730 per second at 720p**, **$0.2205 at 480p**, and **≈$1.04 at 1080p** (1080p rate ⚠️ UNVERIFIED — confirm in the Fal playground before release), so a 30-second 720p clip is about **$14**. On the Reference tier, attaching a video reference applies a **×0.6 multiplier** — but the input video's duration is billed alongside the output's, so a long source clip is not free.
#### Google Omni Flash
Find Omni Flash under **Generate Media → Video → Google → Omni Flash**.
Use Google Omni Flash when you want synchronized audio and video generated in one pass.
Best for:
* Text-to-video and image-to-video with native audio.
* 720p clips from 3–10 seconds.
* 16:9 or 9:16 social and landscape formats.
* Scenes where dialogue, ambient sound, or lip-sync should match the visuals.
**Omni Flash Reference** accepts 2–7 images. When you attach multiple images, the app auto-selects Reference mode.
**Omni Flash Edit** is a video-to-video path under Google alongside Veo Extend. Use it when you want to edit or extend an existing clip while keeping Omni Flash's synchronized audio behavior.
#### Hailuo 03
Use Hailuo 03 when action and motion are the main challenge, or when you want 2K with sound already on it.
Best for:
* Sports.
* Fast movement.
* Stunts.
* Energetic social clips.
* Dynamic camera action.
Every Hailuo 03 tier renders at **2K** — the only resolution offered — for **5 to 15 seconds**, with **native stereo audio always on**. There is no quality tier to choose and no audio toggle to forget.
* **Hailuo 03** (text-to-video): aspect ratios 21:9, 16:9, 4:3, 1:1, 3:4, 9:16, default 16:9. Prompts up to 2000 characters.
* **Hailuo 03** (image-to-video): attach one image to animate it, or two to interpolate a start frame into an end frame. There is no aspect control here — the output follows the source image, so crop the still first if you need a different shape.
* **Hailuo 03 Reference**: up to 9 reference images, 3 reference video clips (2–15s each, 50 MB total), and 3 reference audio clips (2–15s each, 15s combined), to a maximum of **12 files in total**. Audio can never be the only reference. Aspect ratio defaults to **Adaptive**, which follows your references, or pick a concrete ratio.
It is less of a first choice for dialogue-heavy work — reach for Veo 3.1 or Grok Imagine 1.5 when lip-sync matters. Hailuo 03 also takes its time: it renders 2K with audio and can sit in the queue for a while, so send it and carry on editing rather than waiting on it.
#### Wan 2.7
Use Wan 2.7 when you want flexible aspect ratios, high-resolution visual output, or reference-driven generation without generated audio.
Best for:
* Clean visual generations.
* Flexible formats.
* Reference mode with more image inputs.
* Start/end interpolation.
* Instruction-style video editing tests.
Plan to add sound later if the final edit needs audio.
#### Flux 3
Use Flux 3 when you want a long shot at 1080p that already has sound on it.
Best for:
* Clips of **5–20 seconds** at 720p or 1080p.
* Long shots that need native audio in the same pass.
* Ultra-wide framing: Flux 3 offers eight aspect ratios, and **no other video model in the app offers 2:1**.
* Start-frame-to-end-frame transitions that need room to play out.
* 1080p delivery without thinking about quality tiers.
Flux 3 comes from Black Forest Labs and appears in the picker under the **Flux** provider group as simply **Flux 3**.
* **Duration:** 5–20 seconds, whole seconds only, default 8s. Most other families stop at 15 seconds; Seedance 2.5 runs further, to 30s, at up to 1080p.
* **Resolution:** 720p or 1080p, defaulting to **1080p**. There is no 4K tier and no separate quality toggle — resolution *is* the quality setting.
* **Audio:** native and generated with the picture. It ships **On** and, unlike Hailuo 03 or Grok, it can genuinely be switched off. Black Forest Labs recommends describing the ambient sound in your prompt.
* **Aspect ratios:** auto, 21:9, 2:1, 16:9, 4:3, 1:1, 3:4, 9:16. Text-to-video defaults to 16:9; image-to-video defaults to auto but is not locked to the source image.
* **Two images:** attach exactly two and Flux 3 routes to a dedicated first/last-frame endpoint, treating the first as the start frame and the second as the end frame.
{% hint style="warning" %}
Flux 3 has **no reference tier**. Attach one image and only the image-to-video tier is offered — the text-to-video tier hides. Attach three or more images and Flux 3 leaves the picker entirely, handing off to a reference model such as Veo 3.1 Reference. There is also no seed input, no negative prompt, and no 4K.
{% endhint %}
Cost scales purely with duration: **$0.17 per second at 720p** and **$0.29 per second at 1080p**, the same across all three Flux 3 endpoints. A 20-second 1080p clip is **$5.80**, so draft short at 720p and stretch the duration only on the take you are keeping.
#### Luma Ray 3.2
Use Luma Ray 3.2 when you want a silent cinematic clip you will score yourself, a cheap drafting tier, or an edit pass over existing footage with explicit control over how far it strays.
Best for:
* Silent plates and B-roll you plan to sound-design in Premiere.
* Cheap drafting — the **540p tier** prices well below the 720p default.
* A quick, inexpensive start/end interpolation from two stills.
* Restyling or editing an existing clip with **Luma Ray 3.2 Edit** and its divergence dial.
The family appears in the picker under the **Luma** provider group. The generation tiers are labelled **Luma Ray 3.2**; the edit model is **Luma Ray 3.2 Edit** in the video-to-video section, and it is also selectable inside **Studio → Add Effects**.
* **Duration:** 5s or 10s on text-to-video and Edit, default 5s. **Image-to-video is 5s only** — 10s needs multi-keyframe input that Chat Video Pro does not expose.
* **Resolution:** 540p, 720p, or 1080p on every tier, defaulting to **720p**. Resolution is the only quality setting.
* **Audio:** **none, on any tier.** There is no audio pill anywhere in the family — the models have no audio parameters, so nothing can switch sound on. Build the soundtrack in Premiere.
* **Aspect ratios:** 16:9 (default), 9:16, 1:1, 4:3, 3:4, 21:9 — no auto. The Edit model has no aspect setting at all; its output follows the source clip.
* **Two images:** attach exactly two and the second becomes the end frame **on the same image-to-video endpoint** — unlike Flux 3, nothing re-routes to a separate first/last-frame endpoint.
* **The Edit divergence dial:** Default (Luma decides), Adhere (closest to the source), Balanced, Reimagine (diverges the most).
{% hint style="warning" %}
Luma Ray 3.2 has **no reference tier** and narrows the same strict way Flux 3 does: attach one image and only the image-to-video tier is offered; attach three or more images (or use Element tags) and the family leaves the picker entirely. There is also no seed, no negative prompt, and nothing above 1080p. Do not confuse it with **Studio Reframe** — that is a separate Luma-powered Studio surface with its own page, and it never appears in the chat video-to-video picker.
{% endhint %}
Pricing is per second, constant across 5s and 10s, and differs per endpoint: text-to-video **$0.20/s at 720p** ($0.10 at 540p, $0.40 at 1080p), image-to-video **$0.06/s at 720p** ($0.03 at 540p, $0.24 at 1080p), Edit **$0.216/s at 720p** ($0.144 at 540p, $0.432 at 1080p). A 10s text-to-video clip at the default tier is $2.00; a 10s Edit pass is $2.16.
#### Grok Imagine 1.5
Use Grok Imagine 1.5 when speed matters and you want the clip to come back with sound already on it.
Best for:
* Up to **1080p** on both text-to-video and image-to-video.
* Lipsync on single-image animation.
* Fast drafts and social clips.
* Quick content experiments.
Audio is always on across the Grok video tiers — there is no toggle. Duration runs 1 to 15 seconds.
**Grok Reference** accepts up to 7 images for reference-driven generation, at 480p or 720p.
Grok Imagine 1.5 is useful for fast exploration. For final cinematic polish, compare against Veo, Kling, Sora, or Seedance.
***
### Which Mode Should I Use?
Model choice matters, but mode choice comes first.
| You have... | Use... |
| ------------------------------------------------------------- | ---------------------------------------- |
| Only a prompt | Text-to-Video |
| One still image | Image-to-Video |
| A start frame and end frame | Transition Mode or Studio AI Transitions |
| Several images of the same subject | Reference Mode |
| An existing video that should continue | Generative Extend |
| A still image that needs directed camera movement | Studio Motion Director |
| A video that needs cleanup, VFX, relight, reshoot, or upscale | Studio |
If you are not sure, start with the inputs. The number and type of files you attach usually determines the best mode.
***
### Audio Guide
| Need | Recommended models |
| ------------------------------- | ----------------------------------------------------------------------------------- |
| Dialogue or lip-sync | Veo 3.1, Veo 3.1 Fast, Google Omni Flash |
| I2V lipsync from a still | Grok Imagine 1.5 |
| Ambient sound or general audio | Kling 3.0, Kling O3, Seedance 2, Seedance 2.5, Google Omni Flash, Hailuo 03, Flux 3 |
| Audio on a clip longer than 15s | Flux 3 (to 20s, up to 1080p) or Seedance 2.5 (to 30s, up to 1080p) |
| Silent visual draft | Veo 3.1 Lite, Sora 2, Wan 2.7, Seedance 2 Mini, Luma Ray 3.2 |
| Final sound design | Generate visuals first, then finish audio in Premiere |
Audio generation can be useful, but it is not always the best final audio. For client work, you may still want to add dialogue, voiceover, music, or sound effects in Premiere.
***
### Pro vs. Fast vs. Standard
Some model families have quality or speed variants.
Use the higher-quality option when:
* The shot is for delivery.
* The prompt is complex.
* Character or product consistency matters.
* You are generating a final hero shot.
Use the faster or lighter option when:
* You are exploring ideas.
* You are testing prompts.
* You expect to regenerate several times.
* You care more about speed or cost than final polish.
A good workflow is to draft with a faster model, then regenerate the best prompt with a higher-quality model.
***
### When To Use Studio Instead
Studio is better than manual model selection when the task already has a clear workflow.
Goal
Use Studio workflow
Create a cinematic still before video work
Cinematic Lab
Animate a still with a camera move
Motion Director
Create a polished transition
AI Transitions
Generate alternate angles
Multi-Cam
Transfer motion to a character image
Motion Capture
Add VFX to a video
Add Effects
Change lighting
Relight Scene
Reshoot a short segment
Reshoot
Finish resolution
Upscale
Studio does more of the prompt and model setup for you. Direct model selection gives you more control when you already know exactly which generation mode you want.
***
### Practical Starting Points
#### If you are new
Start with one of these:
* **Veo 3.1 Fast** for quick audio-enabled video tests.
* **Kling 3.0 Pro** for cinematic motion and image-to-video.
* **Veo 3.1 Lite** for lower-cost silent drafts.
* **Studio Motion Director** if you already have a strong still image.
#### If you are making B-roll
Try:
* Sora 2 for cinematic visual clips.
* Kling 3.0 or Kling O3 for stronger movement.
* Veo 3.1 Lite for lower-cost drafts.
* Seedance 2 if audio/ambient sound is useful.
#### If you are making social content
Try:
* Grok Imagine 1.5 for fast turnaround, up to 1080p, and I2V lipsync.
* Hailuo 03 for action and energy at 2K with sound.
* Kling for polished camera movement.
* Veo 3.1 when dialogue or sound matters.
#### If you are making client-facing shots
Use faster models to explore, then move the best idea into a quality pass:
1. Draft with a faster or lighter model.
2. Refine the prompt.
3. Regenerate with a higher-quality model.
4. Use Studio tools for cleanup, transitions, relight, or upscale.
***
### Troubleshooting
#### The model I expected is not available
The model selector changes based on what you attach. No attachments show text-to-video models. One image shows image-to-video options. Two images may activate transition mode. Multiple reference images may show reference models.
#### The result has no audio
Check whether the selected model supports audio. Veo 3.1, Kling, Seedance (2 and 2.5 alike), Google Omni Flash, Hailuo 03, and Flux 3 can generate audio — on Hailuo 03 it is always on, with no toggle. On Flux 3 and on the Seedance tiers the audio pill defaults to On but can be switched off, and Chat Video Pro remembers that choice across models, so if you turned audio off on an earlier generation it will still be off here. Grok Imagine 1.5 supports lipsync on image-to-video. Sora 2, Veo 3.1 Lite, and Wan 2.7 are better treated as visual models unless the app shows an audio option for your selected mode. Luma Ray 3.2 is silent on every tier — there is no audio pill in the family at all, so a silent result there is expected, not a bug.
#### The model looks wrong for my use case
Switch based on the failure:
| Problem | Try |
| ------------------------------------ | --------------------------------------------------------------------------- |
| Weak dialogue or lip-sync | Veo 3.1 |
| Weak motion | Kling, Seedance, or Hailuo 03 |
| Clip is too short | Flux 3 (to 20s at 1080p) or Seedance 2.5 (to 30s at up to 1080p) |
| Need faster drafts | Veo 3.1 Lite, Seedance 2 Mini, Grok Imagine 1.5, or a Fast/Standard variant |
| Need stronger transition | Studio AI Transitions |
| Need more controlled image animation | Studio Motion Director |
| Need a cleaner final | Upscale after the creative result is approved |
#### The model list changes over time
Chat Video Pro adds and updates models as providers improve. If the in-app selector differs from this page, trust the app. This guide is meant to help you choose the right kind of model, not memorize every technical option.
***
### Next Steps
* Use [Text-to-Video](/features/video-generation/text-to-video) when starting from a prompt.
* Use [Image-to-Video](/features/video-generation/image-to-video) when animating one still.
* Use [Transition Mode](/features/video-generation/transition-mode) or [AI Transitions](/features/studio/ai-transitions) when connecting two frames.
* Use [Reference Mode](/features/video-generation/reference-mode) when subject consistency matters.
* Use [Studio](/features/studio) for guided production and post-production workflows.
# Text-to-Video
Create videos from text descriptions using AI. Simply describe what you want to see, and Chat Video Pro generates it.
Text-to-Video is the simplest way to create a new AI video from scratch. You write what you want to see, choose a video model, set the basic generation options, and Chat Video Pro generates the shot.
Use it when you do not already have a source image, start frame, end frame, reference set, or existing video. It is best for creating new B-roll, establishing shots, product concepts, social clips, background plates, style tests, and visual ideas that do not exist yet.
{% hint style="info" %}
If you already have a still image, use Image-to-Video or Studio Motion Director. If you have two frames, use Transition Mode or Studio AI Transitions. Text-to-Video is for starting from words.
{% endhint %}
***
### When To Use Text-to-Video
Use Text-to-Video when you want the model to invent the whole shot:
* A quick B-roll insert.
* A cinematic establishing shot.
* A product or brand concept.
* A background plate for an edit.
* A social clip idea.
* A mood, setting, or visual direction test.
* A shot you will later refine with Studio tools.
Choose another workflow when:
You have...
Use instead
One still image to animate
Image-to-Video or Studio Motion Director
Two images to connect
Transition Mode or Studio AI Transitions
Reference images of a character/product
Reference Mode
Existing video to extend
Generative Extend
Existing video to clean up or edit
Studio or Video Canvas Editor
A cinematic still before motion
Studio Cinematic Lab
Help writing a strong prompt
Video Prompter Assistant
***
### How It Works
1. Enable [**Generate Media**](/getting-started/interface-overview/generate-media-button) in the composer.
2. Choose a video model from the model dropdown.
3. Set the aspect ratio, duration, resolution, and audio options.
4. Write a prompt that describes the shot.
5. Generate the video.
6. Review the result.
7. Regenerate, edit, extend, send to Studio, or import into Premiere.
The first generation is often a draft. Treat Text-to-Video like directing a model: generate, review what worked, then refine the prompt or switch models if needed.
***
### What A Good Prompt Includes
A strong Text-to-Video prompt usually answers six questions:
Prompt element
What to describe
Subject
Who or what is the shot about?
Setting
Where and when does it happen?
Action
What changes during the shot?
Camera
How is the shot framed or moving?
Style
What visual language should it follow?
Lighting and mood
What should it feel like?
Useful structure:
{% code overflow="wrap" %}
```
[Camera movement] on [subject] doing [action] in [setting], [lighting], [style], [mood], [important details].
```
{% endcode %}
You do not need to use this exact format every time. The goal is to give the model enough direction to understand the shot, not to write a novel.
***
### Prompt Examples
#### Cinematic B-Roll
{% code overflow="wrap" %}
```
Slow push-in on a steaming coffee cup on a wooden table in a cozy coffee shop during golden hour, warm natural light through large windows, soft bokeh background, peaceful and inviting atmosphere.
```
{% endcode %}
Why it works:
* Clear subject: coffee cup.
* Clear camera move: slow push-in.
* Clear setting and light: cozy coffee shop, golden hour.
* Clear mood: peaceful and inviting.
#### Product Shot
{% code overflow="wrap" %}
```
Smooth orbit around a premium black smartwatch on a dark reflective surface, studio lighting, subtle rim light on the edges, luxury product commercial style, clean background, slow elegant motion.
```
{% endcode %}
Why it works:
* The product is specific.
* The camera move is direct.
* The lighting and style support the use case.
#### Establishing Shot
{% code overflow="wrap" %}
```
Wide aerial shot of a coastal city at sunrise, camera gliding slowly over rooftops toward the ocean, soft golden light, light morning haze, cinematic documentary style, calm optimistic mood.
```
{% endcode %}
Why it works:
* It gives scale and motion.
* The camera has a path.
* The mood fits an establishing shot.
#### Social Clip
{% code overflow="wrap" %}
```
Vertical handheld shot following a runner through a neon-lit city street at night, wet pavement reflections, energetic pacing, quick natural camera movement, modern social ad style.
```
{% endcode %}
Why it works:
* It specifies vertical framing.
* Motion and platform style are clear.
* The scene has visual hooks.
***
### Weak Prompts To Avoid
Weak prompts usually fail because they describe a topic, not a shot.
Too vague:
```
A coffee shop.
```
Better:
{% code overflow="wrap" %}
```
Slow push-in on a barista pouring latte art in a cozy coffee shop, warm morning light, shallow depth of field, cinematic lifestyle ad.
```
{% endcode %}
Missing action:
```
A person in a city.
```
Better:
{% code overflow="wrap" %}
```
Tracking shot following a person crossing a busy city street at sunset, traffic lights glowing, wind moving their coat, cinematic urban energy.
```
{% endcode %}
No camera or style:
```
A car driving.
```
Better:
{% code overflow="wrap" %}
```
Low-angle tracking shot of a vintage red car driving along a coastal highway at sunset, ocean in the background, warm film look, smooth cinematic motion.
```
{% endcode %}
***
### Choosing A Model
You do not need to memorize every model. Choose based on the hardest part of the shot.
| Need | Good starting point |
| ---------------------------------------------- | ------------------------------------- |
| Dialogue or generated audio | Veo 3.1 or Veo 3.1 Fast |
| Audio and video in one pass | Google Omni Flash |
| Lower-cost silent drafts | Veo 3.1 Lite or Seedance 2 Mini |
| Cinematic motion | Kling 3.0 Pro or Kling O3 |
| Longer visual clips without audio | Sora 2 or Sora 2 Pro |
| Natural motion with ambient audio | Seedance 2 |
| Action or sports | Hailuo 03 |
| 2K output with native audio | Hailuo 03 |
| Fast turnaround with audio and lipsync | Grok Imagine 1.5 |
| Flexible visual output without generated audio | Wan 2.7 |
| A clip of 15–20 seconds at 1080p | Flux 3 (up to 20s) |
| A clip that runs past 20 seconds | Seedance 2.5 (up to 30s, up to 1080p) |
| Ultra-wide 2:1 or 21:9 framing | Flux 3 |
| A silent plate with a cheap 540p draft tier | Luma Ray 3.2 |
For a deeper chooser guide, see Supported Video Models.
{% hint style="info" %}
**Flux 3 runs 5–20 seconds with native audio, at 720p or 1080p (1080p by default).** Most families stop at 15 seconds, so if a shot needs to breathe — a long developing camera move, a slow reveal, an unbroken atmospheric hold — Flux 3 carries it at full resolution. Cost scales with duration: $0.17/second at 720p and $0.29/second at 1080p, so a 20-second 1080p clip is $5.80. Draft at 5–8 seconds first.
Flux 3 has no seed and no negative prompt, so steer entirely with positive description — and because audio is generated with the picture, write the soundscape into the prompt.
{% endhint %}
{% hint style="info" %}
**Seedance 2.5 goes further on duration, to 30 seconds, at up to 1080p (default 720p).** It runs **4–30 seconds** with native audio at 480p, 720p, or 1080p, and its default duration is **Auto**, where the model picks the length from your prompt. That makes the Flux 3 / Seedance 2.5 choice a straight trade: Flux 3 for up to 20s, Seedance 2.5 for running time up to 30s — both at 1080p. Seedance 2.5 also ships alongside Seedance 2 rather than replacing it, because Seedance 2's standard tier still adds **4K**.
Multi-shot sequencing on Seedance 2.5 is **prompt-driven** — write the cuts into the prompt ("Cut to a wide shot of…") and one generation can carry several shots. There is no multi-shot control to switch on. Billing is token-based ($0.0214 per 1,000 tokens, `height × width × seconds × 24 ÷ 1024`), so the per-second cost depends on the aspect ratio; at 16:9 that is roughly $0.4730/second at 720p, $0.2205 at 480p, and ≈$1.04 at 1080p (1080p rate ⚠️ UNVERIFIED — confirm in the Fal playground before release), putting a 30-second 720p clip near $14.
{% endhint %}
{% hint style="info" %}
**Luma Ray 3.2 is the silent-by-design option.** It generates 5s or 10s clips at 540p, 720p, or 1080p (default 720p) with no audio on any tier — there is no audio pill to check, because the models have no audio parameters. That makes it a clean choice for plates you will score in Premiere, and the 540p tier is priced for drafting ($0.10/second, against $0.20/second at 720p and $0.40 at 1080p). Like Flux 3 it has no seed and no negative prompt, so steer with positive description — but skip the soundscape, since none will be generated.
{% endhint %}
***
### Settings That Matter
#### Aspect Ratio
Choose the aspect ratio for the edit, not just the idea.
Aspect ratio
Use for
16:9
YouTube, websites, landscape edits, most Premiere timelines.
9:16
TikTok, Reels, Shorts, vertical ads.
1:1
Square social placements when supported by the selected model.
21:9
Cinematic widescreen when supported.
4:3 or 3:4
Vintage, editorial, or portrait alternatives when supported.
If the model does not support the aspect ratio you need, switch models or generate in the closest supported format and reframe in Premiere.
#### Resolution
Generate at the resolution that fits the stage of work.
* Use lower or standard resolution for drafts.
* Use higher resolution for final candidates.
* Use Studio Upscale after the creative result is approved.
#### Audio
Only some models generate audio. If the clip needs speech, sound effects, or ambient audio, choose a model that supports it and describe the audio in the prompt.
If you plan to build the final sound in Premiere, a silent model can be a better choice.
***
### Best Practices
#### Describe A Shot, Not A Concept
The model needs direction. "Luxury watch commercial" is a concept. "Slow orbit around a black watch on a reflective surface with rim lighting" is a shot.
#### Give The Camera Something To Do
Camera movement is one of the strongest levers in video generation. Use phrases like:
* Slow push-in.
* Wide establishing shot.
* Tracking shot.
* Handheld follow shot.
* Low-angle dolly.
* Smooth orbit.
* Aerial glide.
#### Keep The Action Believable
AI video models struggle when too much changes at once. A simple clear action usually beats five competing actions.
Better:
```
The runner turns the corner and accelerates down the wet street.
```
Riskier:
{% code overflow="wrap" %}
```
The runner jumps over a car, changes outfits, enters a building, and the scene becomes a concert.
```
{% endcode %}
#### Mention What Matters Most
If a detail must be right, put it in the prompt. If the car must be red, say red. If the shot must be vertical, set the aspect ratio and mention vertical social ad framing.
#### Use Studio After The First Good Result
Once a Text-to-Video result is close, use Studio to improve it:
Next step
Studio workflow
Add VFX or atmosphere
Add Effects
Change lighting
Relight Scene
Fix a short moment
Reshoot
Remove a distraction
Erase Objects
Improve resolution
Upscale
***
### Example Workflows
#### Quick B-Roll
1. Choose a fast or balanced model.
2. Use a simple prompt with subject, camera, setting, and mood.
3. Generate a 4-8 second clip.
4. Regenerate once or twice with clearer camera/action notes.
5. Import the best result into Premiere.
#### Client Concept Shot
1. Start with a detailed prompt.
2. Draft with a faster model or lower resolution.
3. Refine the wording based on the result.
4. Regenerate with a higher-quality model.
5. Use Studio Upscale or Add Effects only after the creative direction is approved.
#### Social Video Insert
1. Set aspect ratio to 9:16.
2. Choose a model that supports the format you need.
3. Prompt for vertical composition and mobile pacing.
4. Keep the action simple and readable.
5. Finish captions, music, or sound design in Premiere.
***
### Troubleshooting
#### The video does not match my prompt
Make the prompt more concrete. Add a clearer subject, camera move, setting, action, and style. If the model keeps missing the same thing, try a different model.
#### The shot feels generic
Add specific production language:
{% code overflow="wrap" %}
```
35mm handheld documentary style, natural window light, shallow depth of field, soft background motion.
```
{% endcode %}
or:
{% code overflow="wrap" %}
```
Premium product commercial style, slow controlled orbit, black reflective surface, sharp rim light.
```
{% endcode %}
#### The motion is messy
Simplify the action. Ask for one clear movement instead of several. If the shot depends on camera movement, try Kling, Seedance, Hailuo 03, or Studio Motion Director depending on the source.
#### The audio did not generate
Check whether the selected model supports audio and whether the audio toggle is enabled. Veo 3.1, Kling, Seedance (2 and 2.5 alike), Hailuo 03, Grok Imagine 1.5, and Flux 3 are the choices when audio matters — on Hailuo 03 and Grok there is no toggle at all, because audio is always on. Flux 3 does have a toggle: it defaults to On, but Chat Video Pro remembers your last audio choice across models, so it can arrive switched off if you disabled audio on an earlier generation. Veo 3.1 Lite, Sora, and Wan are better treated as visual models unless the app shows audio support for that mode. Luma Ray 3.2 never generates audio on any tier — a silent Luma result is by design.
#### The result is close but not final
Do not keep regenerating blindly. Decide what is wrong:
Problem
Better next step
Needs better prompt
Rewrite with clearer camera/action details.
Needs stronger model
Switch models using Supported Video Models.
Needs VFX or lighting
Use Studio Add Effects or Relight Scene.
Needs cleanup
Use Studio Erase Objects, Rotoscope, or Reshoot.
Needs resolution
Use Studio Upscale after approval.
***
### Related Pages
* [Supported Video Models](/features/video-generation/supported-video-models) - Choose the right model.
* [Image-to-Video](/features/video-generation/image-to-video) - Animate a still image.
* [Transition Mode ](/features/video-generation/transition-mode)- Connect two frames.
* [Reference Mode](/features/video-generation/reference-mode) - Use references for consistency.
* [Video Prompter Assistant](/conversation-starters/video-prompter-assistant) - Get help building a stronger prompt.
* [Studio](/features/studio) - Use guided production and post-production workflows.
***
**Next:** If you already have an image you want to animate, use Image-to-Video or Studio Motion Director.
# Image-to-Video
Animate still images into motion by uploading an image and describing the movement you want. Perfect for bringing static artwork, photos, or designs to life.
Image-to-Video turns a still image into a moving shot. You provide the first frame, then describe how the camera, subject, and environment should move.
Use it when the image already has the visual direction you want: a product photo, generated frame, concept image, character portrait, landscape, frame grab, title-card design, or artwork. The model does not need to invent the entire scene from text. It starts from your image and tries to bring it to life.
{% hint style="info" %}
If your main goal is a controlled camera movement like dolly in, orbit, crane up, tracking shot, or handheld, use Studio Motion Director. It is a guided Image-to-Video workflow built specifically for camera moves.
{% endhint %}
***
### When To Use Image-to-Video
Use Image-to-Video when you already have the frame and need motion:
* Animate a product image.
* Bring a Cinematic Lab frame to life.
* Turn concept art into a short moving insert.
* Add movement to a landscape, environment, or real estate still.
* Animate a character portrait or illustration.
* Create motion from a Premiere frame capture.
* Test whether a still can become usable B-roll.
Choose another workflow when:
You have...
Use instead
Only a text idea
Text-to-Video
A still image and want guided camera presets
Studio Motion Director
Two images to connect
Transition Mode or Studio AI Transitions
Several references for the same character/product
Reference Mode
A still image that needs to be created first
Studio Cinematic Lab
Existing footage to change
Studio
***
### How It Works
1. Enable [**Generate Media**](/getting-started/interface-overview/generate-media-button) in the composer.
2. Attach one image.
3. Choose a model that supports Image-to-Video.
4. Set aspect ratio, duration, resolution, and audio options.
5. Write a motion-focused prompt.
6. Generate the video.
7. Review the motion and regenerate, edit, or send the result into Studio.
The image provides the visual anchor. Your prompt should describe motion, not re-describe the picture.
***
### Start With A Strong Image
The source image matters more than the model. A weak still usually becomes a weak video.
Good Image-to-Video sources have:
* A clear subject.
* Readable composition.
* Enough room for motion.
* A stable aspect ratio that matches the final video.
* Visible depth: foreground, subject, and background.
* Good lighting and texture.
* No important details cut off at the edge.
Risky source images include:
* Tiny subjects in a busy frame.
* Faces or hands already distorted.
* Important text or logos that must remain perfect.
* Flat graphics with no depth.
* Cropped portraits with no room for camera movement.
* Images with multiple subjects when only one should move.
If you need a better still before animating, create one in Studio Cinematic Lab first.
***
### Write Motion Prompts, Not Image Prompts
The image already tells the model what the scene looks like. Your prompt should tell it what changes over time.
Focus on:
Motion layer
What to describe
Camera movement
Push in, pull out, pan, orbit, crane, handheld, tracking.
Useful structure:
{% code overflow="wrap" %}
```
[Camera movement] while [subject action], with [environment motion], preserving [important details], [mood/style].
```
{% endcode %}
***
### Good Prompt Examples
#### Product Animation
{% code overflow="wrap" %}
```
Slow orbit around the product, subtle reflections moving across the surface, soft studio lights gliding in the background, premium commercial feel, keep the product shape and logo stable.
```
{% endcode %}
#### Portrait Motion
{% code overflow="wrap" %}
```
Slow push-in on the subject as they turn slightly toward camera, gentle wind moves their hair and jacket, background lights drift softly out of focus, cinematic portrait mood.
```
{% endcode %}
#### Landscape Motion
{% code overflow="wrap" %}
```
Gentle aerial glide forward over the landscape, clouds drifting slowly, grass and trees moving in a light breeze, warm sunrise atmosphere, calm cinematic pacing.
```
{% endcode %}
#### Artwork Or Poster Motion
{% code overflow="wrap" %}
```
Subtle parallax move into the artwork, foreground elements drifting slightly, soft dust particles in the light, gentle camera push-in, preserve the original composition and style.
```
{% endcode %}
#### Character Illustration
{% code overflow="wrap" %}
```
The character takes one confident step forward, cape and hair moving in the wind, camera slowly pushes in, background glows with subtle atmospheric motion, preserve the character identity.
```
{% endcode %}
***
### Weak Prompts To Avoid
Too static:
```
A person in a coffee shop.
```
Better:
{% code overflow="wrap" %}
```
Slow push-in as the person lifts the coffee cup, steam rising gently, background customers moving softly out of focus, warm morning light.
```
{% endcode %}
Too vague:
```
Make it move.
```
Better:
{% code overflow="wrap" %}
```
Slow dolly forward toward the subject, subtle parallax between foreground table and background window, gentle handheld realism.
```
{% endcode %}
Too much change:
```
Turn this portrait into a city chase scene with explosions and a new outfit.
```
Better:
{% code overflow="wrap" %}
```
The subject turns toward camera with a tense expression, wind moves their coat, city lights flicker behind them, dramatic suspense mood.
```
{% endcode %}
Image-to-Video works best when the prompt respects the still image instead of trying to replace it.
***
### Aspect Ratio And Framing
Match the video aspect ratio to the source image whenever possible.
Source image
Best video aspect ratio
Landscape frame
16:9
Vertical portrait or social frame
9:16
Square design or product post
1:1
Cinematic wide frame
21:9, if supported by the selected model
Why this matters:
* It prevents unexpected cropping.
* It keeps the first frame close to your original image.
* It preserves product, face, and composition placement.
* It gives the model a stable visual anchor.
If the source image is too tight, generate or crop a wider version first. Image-to-Video needs room to move.
***
### Choosing A Model
Use the model based on the hardest part of the animation.
| Need | Good starting point |
| ---------------------------------------------- | -------------------------------------------------- |
| Dialogue or audio from a still | Veo 3.1, Veo 3.1 Fast, or Google Omni Flash |
| Budget visual drafts without audio | Veo 3.1 Lite or Seedance 2 Mini |
| Strong camera movement | Kling 3.0 Pro, Kling O3, or Studio Motion Director |
| Longer cinematic visual clips | Sora 2 or Sora 2 Pro |
| Natural movement with ambient audio | Seedance 2 |
| High-energy action | Hailuo 03 |
| 2K from a still, with audio | Hailuo 03 |
| Fast animation with lipsync | Grok Imagine 1.5 |
| Flexible output without generated audio | Wan 2.7 |
| A clip of 15–20 seconds at 1080p | Flux 3 (up to 20s) |
| A clip that runs past 20 seconds | Seedance 2.5 (up to 30s, up to 1080p) |
| Ultra-wide 2:1 or 21:9 framing | Flux 3 |
| A cheap, silent 5s animation or start/end pair | Luma Ray 3.2 (540p–1080p) |
For a deeper chooser guide, see Supported Video Models.
{% hint style="info" %}
**Flux 3 animates a still for up to 20 seconds** with native audio at 720p or 1080p (1080p by default). Its aspect ratio defaults to Auto, following your source image, but unlike Hailuo 03 it is not locked to the source: you can override with any of eight ratios, including 2:1 and 21:9.
Two things to know before you pick it. First, **once you attach an image, Flux 3 offers only its image-to-video tier — the text-to-video tier hides**, which is stricter than Kling or Hailuo. Remove the image to get text-to-video back. Second, Flux 3 has **no reference tier**, so at three or more images it disappears from the picker entirely and a reference model such as Veo 3.1 Reference takes over.
Cost scales with duration: $0.17/second at 720p, $0.29/second at 1080p. A 20-second 1080p animation is $5.80.
{% endhint %}
{% hint style="info" %}
**Seedance 2.5 animates a still for 4–30 seconds, or lets the model choose.** The duration pill runs 4 to 30 whole seconds plus **Auto**, which is the default and lets Seedance pick the length from your prompt. Output is 480p, 720p, or 1080p (default 720p, with **no 4K**) and carries native audio you can switch off. Attach a second image and it becomes the **end frame on the same endpoint** — nothing re-routes, the way Luma and Hailuo work rather than the way Flux 3 does.
Unlike Flux 3 and Luma, attaching an image does **not** narrow the Seedance group to one tier: at one or two images you still see the main tiers and the Reference tiers together. At three or more images the group narrows to the Reference tiers, and a Seedance 2.5 escalation lands on **Seedance 2.5 Reference**, never on a Seedance 2 tier.
Billing is token-based, so cost per second depends on the aspect ratio: roughly $0.4730/second at 720p 16:9, $0.2205 at 480p, and ≈$1.04 at 1080p 16:9 (1080p rate ⚠️ UNVERIFIED — confirm in the Fal playground before release), which puts a 30-second 720p animation near $14.
{% endhint %}
{% hint style="info" %}
**Luma Ray 3.2 animates a still for exactly 5 seconds, silently.** The image-to-video tier is 5s only (10s needs multi-keyframe input that Chat Video Pro does not expose), renders at 540p, 720p, or 1080p (default 720p), and generates no audio on any tier. Attach a second image and it becomes the **end frame on the same endpoint** — unlike Flux 3, nothing re-routes to a separate first/last-frame endpoint. Luma narrows the same strict way Flux 3 does: with an image attached, only the image-to-video tier is offered, and at three or more images the family leaves the picker (no reference tier). At the 720p default a 5s animation costs $0.30 ($0.06/second; $0.03 at 540p, $0.24 at 1080p).
{% endhint %}
***
### Image-to-Video vs. Studio Motion Director
Both workflows animate still images, but they are not the same.
Use Image-to-Video when...
Use Studio Motion Director when...
You want direct model control.
You want guided camera movement presets.
You already know the exact model to use.
You want the workflow to build the movement prompt.
The motion is custom or unusual.
The motion is a known shot type: dolly, orbit, crane, tracking, handheld.
You want to test several models manually.
You want a more structured, production-style camera move.
Motion Director is usually better for users who think in shot language. Image-to-Video is better when you want direct control in the model selector.
***
### Best Practices
#### Keep The First Move Simple
Start with one clear motion. Add complexity only after the first result works.
Good:
```
Slow push-in with subtle wind and background light movement.
```
Riskier:
{% code overflow="wrap" %}
```
Push in, orbit around, change the outfit, make the background transform, add a crowd, and reveal a new location.
```
{% endcode %}
#### Preserve Important Details
If something must stay stable, say so:
```
Keep the product shape, label, and color stable.
```
```
Preserve the character's face, outfit, and silhouette.
```
This is especially important for products, faces, logos, packaging, and character art.
#### Use Depth
Image-to-Video benefits from images with foreground, subject, and background separation. Depth gives the model something to animate through parallax.
Good sources:
* Product on a table with background props.
* Portrait with lights behind the subject.
* Landscape with foreground plants and distant mountains.
* Interior scene with layers: doorway, subject, background window.
#### Do Not Ask For A New Scene
If the prompt asks for a totally different location, outfit, or subject, the model may fight the image. Use Text-to-Video or generate a new still in Cinematic Lab instead.
#### Review The First And Last Frames
A result can look good in the middle but drift at the beginning or end. Check whether the first frame still matches your image and whether the final frame stayed coherent.
***
### Example Workflows
#### Animate A Cinematic Lab Frame
1. Create a strong still in Cinematic Lab.
2. Use Image-to-Video or Motion Director.
3. Prompt for one clear camera movement.
4. Review identity, framing, and motion.
5. Use Upscale after the creative result is approved.
#### Product Motion Shot
1. Start with a clean product image.
2. Match the output aspect ratio to the source.
3. Prompt for a slow orbit, push-in, or studio-light movement.
4. Preserve product shape, label, and color.
5. Use the result as B-roll or ad footage.
#### Social Image Animation
1. Start with a vertical 9:16 image.
2. Use a model that supports the needed vertical format.
3. Prompt for bold, readable motion.
4. Keep the action simple for mobile viewing.
5. Add captions, music, or voiceover in Premiere.
***
### Troubleshooting
#### The image barely moves
Make the motion more specific. "Slow push-in" or "camera pans left to reveal the background" is stronger than "animate this image."
#### The motion is too extreme
Use gentler language:
```
Subtle camera push-in, minimal subject movement, preserve the original composition.
```
#### The subject changes identity
Add preservation language and simplify the action:
{% code overflow="wrap" %}
```
Preserve the subject's face, outfit, and body shape. Only add subtle head movement and wind in the hair.
```
{% endcode %}
#### The image is cropped
Match the output aspect ratio to the source image. If you need a different format, create a version of the still in that format first.
#### The model ignores part of the prompt
Reduce the number of instructions. Image-to-Video is easier to control when the prompt has one main camera move, one subject action, and one environmental motion.
#### The result is close but needs polish
Use Studio for the next pass:
| Problem | Better next step |
| ----------------------------------- | ---------------- |
| Needs a more controlled camera move | Motion Director |
| Needs a different angle first | Multi-Cam |
| Needs better lighting | Relight Scene |
| Needs VFX or atmosphere | Add Effects |
| Needs resolution | Upscale |
***
### Related Pages
* [Text-to-Video](/features/video-generation/text-to-video) - Generate from a prompt only.
* [Supported Video Models](/features/video-generation/supported-video-models) - Choose the right model.
* [Transition Mode](/features/video-generation/transition-mode) - Connect two frames.
* [Reference Mode](/features/video-generation/reference-mode) - Use references for consistency.
* [Cinematic Lab](/features/studio/cinematic-lab) - Create a better still before animating.
* [Motion Director](/features/studio/motion-director) - Use guided camera movements.
***
**Next:** If you need a directed camera move from your still, use Studio Motion Director. If you need to connect two frames, use Transition Mode or Studio AI Transitions.
# Transition Mode
Create smooth cinematic transitions between two keyframes. Upload a start and end image, and AI generates the motion that connects them. Perfect for scene transitions, morphing effects, and creative e
Transition Mode creates motion between two still images. You attach a start frame and an end frame, then describe how the model should move from the first image to the second.
Use it when you want a quick AI-generated bridge between two frames from the normal Video Generation composer. For guided transition styles like Seamless Morph, Product Reveal, Time Passage, Smoke Reveal, Focus Pull, Whip Pan, or Flying Cam, use Studio AI Transitions.
{% hint style="info" %}
Transition Mode answers: "How should image A become image B?" If you want the two images treated as character references instead of start/end frames, use Reference Mode with a reference model.
{% endhint %}
***
### When To Use Transition Mode
Use Transition Mode when you have two images and want the model to connect them with motion:
* A before/after reveal.
* A product transformation.
* A day-to-night change.
* A scene-to-scene bridge.
* A mood or lighting change.
* A simple morph between similar subjects.
* A quick transition test from the standard composer.
Choose another workflow when:
You have...
Use instead
One still image to animate
Image-to-Video or Studio Motion Director
A polished transition style in mind
Studio AI Transitions
Multiple images of the same character/product
Reference Mode
Only a text idea
Text-to-Video
Existing footage to transition from/to
Premiere editing tools or Studio post-production workflows
***
### How It Works
1. Enable **Generate Media** in the composer.
2. Attach exactly two images.
3. Chat Video Pro activates Transition Mode.
4. Choose a transition-capable model.
5. Set duration, resolution, and audio options when available.
6. Write a prompt describing the motion between frames.
7. Generate and review the result.
The first image is the starting frame. The second image is the target frame. Your prompt should describe the path between them.
***
### Transition Mode vs. Studio AI Transitions
Both workflows connect two frames, but they are designed for different levels of guidance.
Use classic Transition Mode when...
Use Studio AI Transitions when...
You want the quick composer path.
You want a guided transition workflow.
You already know which model to use.
You want built-in transition styles and better prompt structure.
You want to write the transition prompt yourself.
You want style presets like Product Reveal, Time Passage, Smoke Reveal, or Whip Pan.
You are testing a simple bridge.
You are making a polished transition for an edit.
The practical rule: use Transition Mode for quick direct control, and use Studio AI Transitions for intentional, styled transitions.
***
### Start With Strong Frames
The two images matter more than the model. A transition can only be as clear as the relationship between the start and end frames.
Strong frame pairs usually have:
* Matching aspect ratios.
* A clear visual relationship.
* Similar framing or a transition style that explains the framing change.
* A readable subject in both frames.
* Enough shared structure for the model to connect them.
* A clear answer to "what is changing?"
Weak frame pairs include:
* Different aspect ratios.
* Totally unrelated images.
* One extreme close-up and one extreme wide with no transition logic.
* Two frames packed with small text or UI.
* Different subjects with no shared shape, location, or idea.
* Frames where the subject appears in completely different screen positions.
If the two frames do not naturally relate, generate a better pair first in Cinematic Lab, Multi-Cam, or another image workflow.
***
### Aspect Ratio Rule
Use matching aspect ratios.
Start frame
End frame
Result
16:9
16:9
Best for landscape transitions.
9:16
9:16
Best for vertical transitions.
1:1
1:1
Best for square transitions.
16:9
9:16
Risky. Cropping or distortion likely.
Different ratios
Different ratios
Avoid unless you intentionally want reframing artifacts.
Matching ratios keep the model focused on the transition instead of solving a crop problem.
***
### Write Transition Prompts, Not Image Captions
The images already describe the starting and ending visuals. Your prompt should describe the movement between them.
Focus on:
Prompt layer
What to describe
Transformation
Morph, reveal, wipe, time passage, dissolve, object change.
Useful structure:
{% code overflow="wrap" %}
```
[Type of transition] from the first frame to the second, using [camera/motion path], keeping [important subject] stable, [style/pacing].
```
{% endcode %}
***
### Prompt Examples
#### Product Reveal
{% code overflow="wrap" %}
```
The closed product box opens into the final hero product shot, with a slow cinematic push-in, soft studio light sweep, subtle haze, and premium commercial pacing.
```
{% endcode %}
#### Day To Night
{% code overflow="wrap" %}
```
Gradual time-lapse transition from the daytime street to the night version, lights turning on across the buildings, sky darkening smoothly, camera position staying stable.
```
{% endcode %}
#### Before And After
{% code overflow="wrap" %}
```
Smooth before-and-after transformation where the messy room becomes clean and organized, objects shifting naturally into place, camera locked off, satisfying gradual reveal.
```
{% endcode %}
#### Scene Change
{% code overflow="wrap" %}
```
A fast whip pan hides the cut between the first location and the second, motion blur briefly fills the frame, then lands cleanly on the final image.
```
{% endcode %}
#### Object Morph
{% code overflow="wrap" %}
```
The small seedling grows and transforms into the mature tree shown in the second frame, time-lapse motion, roots and branches expanding naturally, smooth organic pacing.
```
{% endcode %}
***
### Weak Prompts To Avoid
Too vague:
```
Transition between two images.
```
Better:
{% code overflow="wrap" %}
```
Smooth cinematic morph from the first portrait to the second, preserving the face position while clothing and lighting transform gradually.
```
{% endcode %}
Captions the images instead of the motion:
```
A coffee shop and then a beach.
```
Better:
{% code overflow="wrap" %}
```
Camera pushes through the coffee shop window into a bright beach scene, light blooms across the frame, ending on the final beach image.
```
{% endcode %}
No transition logic:
```
Make these connect.
```
Better:
{% code overflow="wrap" %}
```
A foreground object wipes across the lens, briefly covering the frame, then reveals the second scene with matching camera direction.
```
{% endcode %}
***
### Choosing A Model
Use the model based on the hardest part of the transition.
Need
Good starting point
Best morphing quality
Kling O3 Transition or Studio AI Transitions
Audio-enabled transition
Veo 3.1, Kling, or Seedance 2
Lower-cost transition draft
Veo 3.1 Lite (8s)
Fast transition test
Fast or Standard model variants
High-resolution visual interpolation
Wan 2.7
A transition of 15–20 seconds at 1080p
Flux 3 (up to 20s)
A transition that runs past 20 seconds
Seedance 2.5 (up to 30s, 480p, 720p, or 1080p)
A slow, multi-stage transformation
Flux 3 or Seedance 2.5 — room to let the morph play out
A quick, inexpensive silent interpolation
Luma Ray 3.2 (5s)
Strong style presets
Studio AI Transitions
{% hint style="info" %}
**Flux 3 transitions run up to 20 seconds at 720p or 1080p.** Kling transitions stop at 15s and Veo at 8s, so reach for Flux 3 when the transformation is slow or complex **and** you want the higher resolution. If the morph needs to run even longer, Seedance 2.5 goes to 30s at 480p, 720p, or 1080p.
There is no separate "Flux 3 Transition" entry in the picker. Select **Flux 3** image-to-video and attach exactly two images; the app routes the job to a **dedicated first/last-frame endpoint**, treating the first image as the start frame and the second as the end frame. Duration is always explicit on this path — you pick a whole number from 5 to 20 seconds, and the app defaults to 8s.
Cost check before going long: $0.17/second at 720p and $0.29/second at 1080p, so a 20-second 1080p transition is $5.80. Get the frame pairing right at 5s and 720p first, then stretch the duration.
{% endhint %}
{% hint style="info" %}
**Seedance 2.5 interpolations run 4–30 seconds and stay on one endpoint.** Select **Seedance 2.5** and attach exactly two images: the first becomes the start frame, the second the end frame. Like Luma — and unlike Flux 3 — nothing re-routes, because the Seedance 2.5 image-to-video endpoint takes the end frame natively. There is no separate transition entry in the picker.
Duration is 4 to 30 whole seconds, or **Auto** (the default), where the model picks the length from your prompt. The output carries native audio you can switch off, at 480p, 720p, or 1080p — **there is no 4K on 2.5**. Seedance 2 and Seedance 2 Fast still cover the same job at 4–15s if you need 4K. Billing is token-based, roughly $0.4730/second at 720p 16:9, $0.2205 at 480p, and ≈$1.04 at 1080p 16:9 (1080p rate ⚠️ UNVERIFIED — confirm in the Fal playground before release), so a 30-second 720p transition is about $14 — get the frame pairing right at 480p and a short duration first.
{% endhint %}
{% hint style="info" %}
**Luma Ray 3.2 interpolations are 5 seconds, silent, and stay on one endpoint.** Select **Luma Ray 3.2** image-to-video and attach exactly two images: the first becomes the start frame, the second the end frame. Unlike Flux 3, nothing re-routes — Luma's image-to-video endpoint takes the end frame natively, so the second image changes the payload, not the route. The image-anchored path offers no 10s option, and the output carries no audio, so score it in Premiere. At the 720p default a 5s interpolation costs $0.30 ($0.06/second; 540p and 1080p tiers also available).
{% endhint %}
For most users, Studio AI Transitions is the better path when the transition matters to the edit. Classic Transition Mode is best when you want the fastest manual model-selector route.
***
### Best Practices
#### Treat Images Like First And Last Frames
The start frame should look like the first frame of the shot. The end frame should look like the final landing frame. Do not treat them as loose mood references.
#### Keep The Relationship Clear
The model needs to know what changes and what stays. If both images share a subject, location, pose, or composition, the transition usually improves.
#### Use The Same Composition When Possible
Before/after, product reveals, time passage, and object transformations work best when the core subject stays in a similar screen position.
#### Pick A Transition That Explains The Cut
If the images are similar, a morph or time-passage transition can work. If they are different locations, use a camera move, wipe, whip pan, or reveal that hides the change.
#### Avoid Too Much Text
Generated transitions can distort small text, signs, UI, labels, and detailed typography. If text must remain perfect, consider doing the transition in Premiere or generating a clean background transition first.
***
### Example Workflows
#### Quick Composer Transition
1. Prepare two images with the same aspect ratio.
2. Attach both images in Generate Media.
3. Choose a transition-capable model.
4. Write a short motion prompt.
5. Generate and review.
#### Polished Studio Transition
1. Prepare intentional start and end frames.
2. Open Studio AI Transitions.
3. Add the start and end frames.
4. Choose a transition style.
5. Add optional notes.
6. Generate and compare results.
#### Cinematic Frame Pair
1. Create two stills in Cinematic Lab.
2. Keep the same aspect ratio.
3. Make sure the frames have a clear relationship.
4. Use Transition Mode for a quick bridge, or AI Transitions for a guided style.
***
### Troubleshooting
#### Transition Mode does not activate
Make sure exactly two images are attached and Generate Media is enabled. If you attach more than two images, the app may route you toward reference-style workflows instead.
#### The transition crops or distorts
Check that both images have the same aspect ratio. If they do not, crop or regenerate one image before trying again.
#### The result feels like a dissolve
Add a clearer motion path: whip pan, push through, object wipe, smoke reveal, time-lapse shift, focus pull, or morph.
#### The subject changes too much
Use more preservation language:
{% code overflow="wrap" %}
```
Keep the main subject centered and preserve the face shape while only the lighting and wardrobe change.
```
{% endcode %}
#### The two images do not connect
The frame pair may be too unrelated. Create an intermediate frame, use a stronger transition style in AI Transitions, or generate a better start/end pair with Cinematic Lab.
#### You wanted Reference Mode instead
If the images are references for the same subject, choose Reference Mode and select a reference-capable model. Transition Mode treats two images as start and end frames.
***
### Related Pages
* [AI Transitions ](/features/studio/ai-transitions)- Guided transition styles for polished Studio workflows.
* [Image-to-Video](/features/video-generation/image-to-video) - Animate one still image.
* [Reference Mode](/features/video-generation/reference-mode) - Use multiple images as identity/style references.
* [Cinematic Lab](/features/studio/cinematic-lab) - Generate better start/end frames.
* [Supported Video Models](/features/video-generation/supported-video-models) - Choose the right model.
***
**Next:** If you want guided transition styles instead of writing the motion manually, use Studio AI Transitions.
# Reference Mode
Generate videos with consistent characters using 1-9 reference images. Perfect for maintaining character appearance across multiple shots, creating talking head videos, or ensuring visual consistency
Reference Mode generates video using one or more images as visual anchors. Instead of asking the model to invent a character, product, outfit, location, or style from text alone, you provide images that show what should stay consistent.
Use it when identity matters: a recurring character, presenter, product, mascot, brand object, wardrobe, location, or visual look.
{% hint style="info" %}
If you attach exactly two images and the app switches to Transition Mode, that means Chat Video Pro is treating them as a start frame and end frame. To use the images as references instead, choose a reference-capable model such as Kling O3 Reference, Seedance 2 Reference, Seedance 2.5 Reference, Wan 2.7 Reference, Veo 3.1 Reference, Google Omni Flash Reference (2–7 images), Grok Reference (up to 7 images), or Hailuo 03 Reference (up to 9 images, plus reference video and audio clips).
{% endhint %}
***
### When To Use Reference Mode
Use Reference Mode when you want a generated video to follow visual examples:
* A presenter should look like the same person across shots.
* A product should keep its shape, color, and design.
* A character should remain recognizable.
* A location, wardrobe, or brand style should carry across generations.
* You need multiple videos that feel like the same campaign.
* Text-only prompting is not enough to preserve the subject.
Choose another workflow when:
You have...
Use instead
One image that should become the first frame of a video
Image-to-Video or Studio Motion Director
Two images that should connect as start/end frames
Transition Mode or Studio AI Transitions
A still image that needs alternate camera angles
Studio Multi-Cam
A cinematic still that needs to be created first
Studio Cinematic Lab
Existing video to edit
Studio
***
### How It Works
1. Enable **Generate Media**.
2. Attach reference images of the same subject, product, character, or look.
3. Choose a reference-capable model.
4. Write a prompt describing the scene, action, camera, and mood.
5. Configure duration, aspect ratio, resolution, and audio options when available.
6. Generate the video.
7. Review whether the subject stayed consistent.
The references provide identity and visual direction. The prompt provides the new scene and action.
***
### Reference Mode vs. Transition Mode
This is the most common point of confusion.
If the images are...
Use...
The first and last frame of a shot
Transition Mode
Examples of the same person/product/character
Reference Mode
Two frames you want to connect with a polished style
Studio AI Transitions
A single still you want to animate
Image-to-Video or Motion Director
Transition Mode asks: **How should image A become image B?**
Reference Mode asks: **What should stay consistent while the model creates a new shot?**
***
### Choose Strong Reference Images
Good references are clear, consistent, and useful.
Use images that show:
* The same person, product, or character.
* A clear face, silhouette, product shape, or key design detail.
* Different useful angles when possible.
* Similar identity even if pose, expression, or lighting changes.
* Enough resolution for the model to read details.
* The most important visual traits you want preserved.
Avoid references that are:
* Blurry, dark, or low quality.
* Different people or different products.
* Contradictory styles.
* Extreme angles only.
* Heavily filtered or distorted.
* Full of unrelated background clutter.
* Too many images that fight each other.
More references are not always better. A small set of clean references often works better than a large set of mixed-quality images.
***
### How Many References To Use
Use the smallest set that explains the subject.
Reference count
Best for
1 image
Simple product, logo-like subject, or one clear character anchor.
2-3 images
Most people, products, presenters, and brand subjects.
4-7 images
Complex characters, varied angles, or stronger identity preservation.
8-9 images
When the selected model supports it and you have genuinely useful angle/style variety.
If results ignore your references, try better references before adding more. If results copy the references too literally, reduce the count or use more varied images.
***
### What To Prompt
Do not spend the whole prompt describing the character if the reference images already show them. Use the prompt to describe the new shot.
Focus on:
Only if the selected model supports generated audio.
Preservation
What must stay consistent: face, outfit, product shape, logo, color.
Useful structure:
{% code overflow="wrap" %}
```
The referenced subject [action] in [setting], [camera movement/composition], [lighting/style], preserving [identity/product details].
```
{% endcode %}
***
### Prompt Examples
#### Presenter Video
{% code overflow="wrap" %}
```
The referenced presenter speaks directly to camera in a clean modern studio, confident and friendly delivery, medium shot, soft key light, subtle background blur, preserve the face and outfit from the references.
```
{% endcode %}
#### Character Scene
{% code overflow="wrap" %}
```
The referenced character walks through a neon-lit alley at night, looking over their shoulder with suspicion, slow handheld tracking shot, cinematic noir lighting, preserve the character's face, jacket, and silhouette.
```
{% endcode %}
#### Product Clip
{% code overflow="wrap" %}
```
The referenced product sits on a dark reflective surface while the camera slowly orbits, studio rim light catching the edges, premium commercial style, preserve the product shape, color, and label.
```
{% endcode %}
#### Brand Mascot
{% code overflow="wrap" %}
```
The referenced mascot waves from a bright trade show booth, cheerful expression, slow push-in, colorful brand environment, preserve the mascot proportions and costume details.
```
{% endcode %}
***
### Weak Prompts To Avoid
Only describes the reference:
```
A person with brown hair and a blue jacket.
```
Better:
{% code overflow="wrap" %}
```
The referenced person walks confidently through a modern office lobby, natural daylight, medium tracking shot, preserve their face and blue jacket.
```
{% endcode %}
Too vague:
```
The character does something cool.
```
Better:
{% code overflow="wrap" %}
```
The referenced character steps out of a sports car at night, camera low and close, neon reflections on wet pavement, confident cinematic reveal.
```
{% endcode %}
Conflicts with the reference:
```
Make the referenced black backpack into a red suitcase.
```
Better:
{% code overflow="wrap" %}
```batch
Show the referenced black backpack on a clean studio pedestal, slow orbit, premium product lighting, preserve the backpack shape and material.
```
{% endcode %}
If you want to change the subject itself, use image editing or another workflow. Reference Mode is strongest when references are meant to remain recognizable.
***
### Choosing A Model
Choose based on what matters most.
Need
Good starting point
Dialogue or talking head with references
Veo 3.1 Reference
Flexible duration and strong reference quality
Kling O3 Reference
More reference images and no generated audio
Wan 2.7 Reference
Reference video with native audio/ambient sound
Seedance 2 Reference
A reference clip that needs to run past 15 seconds
Seedance 2.5 Reference — 4–30s (or Auto) at 480p, 720p, or 1080p, with native audio
A reference set larger than 12 files
Seedance 2.5 Reference — the model accepts up to 50 references combined across images, videos, and audio, with no per-type caps
2–7 references with synchronized audio
Google Omni Flash Reference
Up to 7 references for fast I2V with audio
Grok Reference
Reference material that is a video clip or an audio clip, not a still
Hailuo 03 Reference — up to 9 images, 3 video clips (2–15s each, 50 MB total), and 3 audio clips (2–15s each, 15s combined), 12 files in total. Audio can never be the only reference
Guided character/product still creation before video
Studio Cinematic Lab
For a broader model chooser, see Supported Video Models.
{% hint style="info" %}
**Seedance 2.5 Reference — two numbers, both true.** The composer shows **9 image slots, 3 video slots, and 3 audio slots**, the same layout every Seedance reference tier uses. The **model itself accepts up to 50 references combined** across all three types, with no per-type caps. That larger budget is headroom at the API rather than something the current composer can fill, so plan around the 9 / 3 / 3 slots you can actually see.
Two rules to keep in mind. **Audio can never be the only reference** — attach at least one image or video alongside it. And **a video reference changes the bill**: the per-second rate drops by a ×0.6 multiplier, but the input video's duration is billed as well as the output's, so a long source clip is not free. Cite references in the prompt as `[Image1]`, `[Video1]`, `[Audio1]`.
Seedance 2 Reference is the tier to use when you need **4K**: 2.5 has no 4K tier, and the Seedance 2 standard reference tier does not.
{% endhint %}
***
### When To Use Studio Instead
Studio is often better when the task has a more specific creative shape.
| Goal | Better Studio workflow |
| ----------------------------------------- | ---------------------- |
| Create a consistent cinematic still first | Cinematic Lab |
| Animate a still with a camera move | Motion Director |
| Create alternate angles of a subject | Multi-Cam |
| Transition between two intentional frames | AI Transitions |
| Transfer motion to a character image | Motion Capture |
Use Reference Mode when you want direct model control from the composer. Use Studio when you want the workflow to guide the prompt, model, and asset setup.
***
### Best Practices
#### Build A Small Reference Set
For recurring people, products, or characters, keep 3-5 strong images in your Library or Recents. Reuse the same set across generations for more consistent results.
#### Keep References Focused
Do not mix unrelated styles unless style mixing is the goal. A product render, a blurry phone photo, and a stylized illustration may confuse the model if they are all meant to define the same subject.
#### Prompt The Scene, Not The Biography
References handle appearance. Your prompt should direct the new shot: where the subject is, what they are doing, how the camera moves, and what the mood is.
#### Preserve What Matters
If a detail must stay consistent, name it:
```
Preserve the product label, black color, rounded silhouette, and silver zipper.
```
```
Preserve the face, hairstyle, red jacket, and slim silhouette.
```
#### Expect Some Drift
Reference Mode improves consistency, but it is not a perfect identity lock. For critical brand, legal, or celebrity likeness work, review carefully and use manual finishing where needed.
***
### Example Workflows
#### Consistent Presenter Clip
1. Attach 2-4 clear images of the presenter.
2. Choose a model that supports audio if the presenter should speak.
3. Prompt the setting, delivery, camera framing, and dialogue.
4. Review face consistency and lip-sync.
5. Finish sound and edits in Premiere.
#### Product Campaign Shot
1. Attach 2-3 clean product references.
2. Prompt a new commercial scene or product movement.
3. Preserve shape, color, label, and material.
4. Generate multiple versions with different camera or lighting direction.
5. Upscale or edit the best result if needed.
#### Character Series
1. Build a small reference set for the character.
2. Reuse it across each shot.
3. Change the prompt for each scene, action, and camera move.
4. Keep wardrobe and core details consistent unless the story requires a change.
***
### Troubleshooting
#### Reference Mode does not activate
Make sure Generate Media is enabled, images are attached, no video is attached, and a reference-capable model is selected. If exactly two images trigger Transition Mode, manually switch to a reference model.
#### The character does not look consistent
Use clearer references, add more useful angles, and include preservation language in the prompt. Avoid mixing references that show different people, outfits, or styles unless that variation is intentional.
#### The output copies the reference too closely
Use fewer references or add more scene/action detail. The model may be treating your references as the whole shot instead of the identity anchor.
#### The result ignores the scene prompt
Your references may be too dominant or too visually similar. Reduce the reference count and make the prompt more specific about setting, action, and camera.
#### The wrong mode activates
Two images often route to Transition Mode. If you want a start/end transition, stay there. If you want identity references, choose Reference Mode manually with a compatible reference model.
#### The duration or audio options are not what you expected
Reference models have different limits. Some support audio, some do not. Some have fixed duration. Choose the model based on the shot's hardest requirement: audio, duration, reference count, or visual quality.
***
### Related Pages
* [Supported Video Models](/features/video-generation/supported-video-models) - Choose the right model.
* [Transition Mode](/features/video-generation/transition-mode) - Connect two frames as start/end images.
* [Image-to-Video](/features/video-generation/image-to-video) - Animate one source image.
* [Cinematic Lab](/features/studio/cinematic-lab) - Create consistent cinematic source frames.
* [Multi-Cam](/features/studio/multi-cam) - Generate alternate angles.
* [Motion Director](/features/studio/motion-director) - Animate a still with guided camera movement.
***
**Next:** If your two images are meant to become a start and end frame, use Transition Mode. If you want a guided transition style, use Studio AI Transitions.
# Video Canvas Editor
The Video Canvas Editor is the central hub for all video editing operations in Chat Video Pro. Access all the editing tools from one unified interface.
The Video Canvas Editor is the classic full-screen editor for working from an existing video result. Use it when you already have a generated or imported clip in Chat Video Pro and want to open a familiar editor surface for rotoscoping, object removal, effects, reshooting, or upscaling.
{% hint style="info" %}
**Use the Video Canvas Editor when you are already looking at the clip you want to edit.** Use \[Studio]\(../studio/README.md) when you want to start from a guided workflow card like Rotoscope, Erase Objects, Add Effects, Reshoot, Upscale, or Relight Scene.
{% endhint %}
***
### What This Page Is For
The Video Canvas Editor is not a separate creative category. It is an entry point.
Use it when:
* A video result is already in chat.
* You imported a clip and want to work from its thumbnail.
* You want to test a quick edit without opening Studio first.
* You want to compare before/after results in the same editor.
* You are comfortable choosing the editing mode manually.
Use Studio instead when:
* You know the job before choosing a clip.
* You want the workflow to open with the correct model and controls already selected.
* You are starting from Premiere timeline media.
* You want the newer guided Studio path for production or post-production tasks.
***
### How To Open It
1. Generate or import a video in Chat Video Pro.
2. Click **Edit** on the video thumbnail.
3. The Video Canvas Editor opens full screen.
4. Choose the editing mode from the selector.
5. Configure the mode-specific controls.
6. Process the edit, review the result, then click **Done** when you want to send it back to chat.
If you are starting from a clip in Premiere, Studio is often faster because the workflow asset loader can pull from your timeline and open the correct tool directly.
***
### Video Canvas Editor vs. Studio
Both paths can reach many of the same underlying AI tools. The difference is how you start.
Start here
Use when
Video Canvas Editor
You already have a video in chat and want to edit that exact result.
Studio
You know the task first and want a guided workflow with the right loader, model, and controls.
Examples:
Goal
Better starting point
Open a generated clip and quickly upscale it
Video Canvas Editor
Pull a selected Premiere timeline clip into Upscale
Studio Upscale
Try an effect on the video result you just made
Video Canvas Editor
Start a clean Add Effects workflow from a source clip
Studio Add Effects
Rotoscope a clip already visible in chat
Video Canvas Editor
Start a guided subject-isolation workflow
Studio Rotoscope
The practical rule: **Video Canvas Editor is clip-first. Studio is task-first.**
***
### Editing Modes
The available modes can vary by release, clip type, and model availability, but the classic editor is mainly used for these video tools:
Mode
Best for
Rotoscope
Selecting and isolating a subject from the background.
Erase Objects
Removing unwanted people, objects, gear, text, or distractions from short clips.
Add Effects
Adding rain, fire, fog, atmosphere, style, or visual transformations.
Reshoot
Regenerating a selected segment of video, audio, or both.
Upscale
Increasing resolution and improving the finish after the edit is approved.
***
### Interface Overview
#### Video Player
Use the player to review the source clip, scrub the timeline, and inspect the result after processing.
#### Model Selector
The selector switches between editing modes. When you change modes, the right-side controls change with it.
#### Mode Controls
Each mode has its own controls:
* Rotoscope uses text, box, or point selection.
* Erase Objects uses text or point selection to mark what should disappear or be protected.
* Add Effects uses a prompt and optional reference images.
* Reshoot uses a timeline segment, prompt, and retake mode.
* Upscale uses a model, scale factor, codec, and quality settings.
#### Before/After Review
After a result is generated, use the before/after view to check whether the edit improved the clip. If the edit missed the target, revise the prompt, selection, segment, or mode before sending it back to chat.
***
### Common Workflows
#### Quick Follow-Up After Generation
1. Generate a video.
2. Click **Edit** on the result.
3. Choose the tool that matches the problem.
4. Run the edit.
5. Compare before/after.
6. Click **Done** when the result is worth keeping.
This is the best use of the Video Canvas Editor: one more pass on a clip that is already in front of you.
#### Rotoscope A Clip From Chat
1. Click **Edit** on the video.
2. Choose the Rotoscope/SAM 3 mode.
3. Select the subject with text, box, or point controls.
4. Track the frame first if available.
5. Track the full video.
6. Remove the background and send the result back.
For the full workflow, use Rotoscope.
#### Add An Effect To An Existing Result
1. Click **Edit** on the video.
2. Choose Add Effects/Kling VFX.
3. Write the effect prompt.
4. Optional: add reference images.
5. Generate and compare the result.
Good prompts describe what changes and what should remain stable. For example: `Add heavy rain and wet street reflections while keeping the same camera move, subject, and composition.`
For deeper effect prompting, use Add Effects.
#### Reshoot One Moment
1. Click **Edit** on the video.
2. Choose Reshoot/LTX Retake.
3. Select the segment you want to change.
4. Describe the replacement action, visual, audio, or both.
5. Generate and compare.
For best results, keep the requested change local and specific. Use Reshoot when you need the complete guided explanation.
#### Upscale A Final Clip
1. Click **Edit** on the video.
2. Choose Upscale.
3. Select the upscaler and scale factor.
4. Choose codec and quality settings if available.
5. Generate the upscaled result.
Do this after the creative edit is approved. Upscaling drafts wastes time and credits.
***
### Best Practices
#### Choose The Tool From The Problem
Do not pick a model first. Ask what is wrong with the clip:
Problem
Use
The background needs to become transparent
Rotoscope
Something should disappear
Erase Objects
The whole shot needs new atmosphere or effects
Add Effects
One moment needs a different action or audio
Reshoot
The final result needs more resolution
Upscale
#### Keep Source Clips Short
Video editing models work best on focused clips. Trim around the moment you want to change instead of sending a long sequence with several unrelated actions.
#### Preview Before Processing When Possible
For selection-based tools, test the first frame or selection before processing the whole clip. A clean selection saves more time than guessing.
#### Upscale Last
Finish creative changes first, then upscale the version you actually want to keep.
***
### Troubleshooting
#### The Edit Button Is Missing
Make sure the item is a video result or imported video. Image results use image workflows instead of the Video Canvas Editor.
#### The Wrong Modes Are Showing
The available modes depend on the source media, model availability, and current app version. If you know the exact workflow you want, open it from Studio instead.
#### The Result Does Not Match The Prompt
Make the instruction more specific. Describe the subject, what should change, what should stay the same, and the intended style or mood. For Reshoot, make sure the selected segment matches the moment described in the prompt.
#### The Tool Fails On A Long Clip
Try a shorter clip or smaller segment. Some models have duration, resolution, or codec constraints. Studio workflow pages list the important limits for each tool.
#### I Am Not Sure Whether To Use This Or Studio
Use the Video Canvas Editor when the clip is already in chat. Use Studio when the task comes first.
***
### Related Pages
* [Video Generation](/features/video-generation) - Create, animate, transition, reference, extend, and edit videos.
* [Studio](/features/studio) - Guided production and post-production workflows.
* [Rotoscope](/features/studio/sam-3-rotoscoping) - Isolate a subject and remove the background.
* [Erase Objects](/features/studio/object-eraser-tool) - Remove unwanted objects from short clips.
* [Add Effects](/features/studio/kling-vfx) - Add VFX, weather, atmosphere, and style.
* [Reshoot](/features/studio/reshoot) - Regenerate a selected section of video or audio.
* [Upscale](/features/studio/video-upscaling) - Increase resolution after the edit is approved.
***
**Next:** If you are starting from a specific task, open Studio. If you already have a clip in chat, click **Edit** and use the Video Canvas Editor.
# Generative Extend
Extend any video by 7 seconds using AI continuation. Perfect for lengthening clips, adding more content to existing videos, or creating seamless extensions.
Generative Extend continues an existing video by 7 seconds. Use it when a clip is working, but it ends too soon and you want the action, camera move, or scene energy to keep going naturally.
This is different from generating a new video. Extend uses your current clip as the source and asks the model to continue from the ending.
{% hint style="info" %}
**Think of Generative Extend as continuation, not replacement.** It is best when the current clip is already close and you want more of it. If the scene, subject, framing, or style is wrong, generate a better clip first.
{% endhint %}
***
### What This Tool Is For
Use Generative Extend when you want to:
* Add time to a short generated clip.
* Continue a camera move that ends too early.
* Let an action finish more naturally.
* Create a longer establishing shot.
* Add breathing room before a cut.
* Stretch a useful result toward a social or edit duration.
Do not use it when you want a completely different scene. Extend is strongest when the next 7 seconds should feel like a natural continuation of the existing video.
***
### When To Use It
Goal
Use Generative Extend?
The shot is good but too short
Yes. This is the ideal use case.
The camera move stops before the reveal
Yes. Prompt the camera to keep moving.
The character should finish the action
Yes. Describe the next action.
You need a new unrelated b-roll shot
No. Use Text-to-Video or Image-to-Video.
The clip is already over 23 seconds
No. Trim first or use another workflow.
You need final 4K output immediately
No. Extend outputs 720p, then upscale later if needed.
The practical rule: extend when the clip deserves more time. Regenerate when the clip needs a different idea.
***
### How To Use It
1. Attach or import a video into Chat Video Pro.
2. Enable **Generate Media** if needed.
3. Choose **Extend Generation** from the video model selector.
4. Optional: write a continuation prompt.
5. Choose whether audio generation is on or off.
6. Generate the extension.
7. Review the result and bring it into Premiere if it works.
Generative Extend is part of the Video Generation bucket because it continues an existing video directly from the composer. It is not a Studio workflow card.
***
### Controls And Constraints
Generative Extend currently uses **Veo 3.1 Extend** through Fal.ai.
Setting
Current behavior
Input
1 video
Input duration
1-23 seconds
Extension duration
Fixed at +7 seconds
Max total output
30 seconds
Output resolution
720p
Supported aspect ratios
16:9 or 9:16
Max file size
200MB
Prompt
Optional, up to 1000 characters
Audio
On by default, can be turned off
If the input is 1080p, the extended output is still 720p. If the final clip needs more resolution, extend first, approve the result, then use Studio Upscale.
***
### Writing Extension Prompts
The prompt should describe what happens next, not everything that already happened.
Good extension prompts usually include:
Prompt element
Example
Continued action
The runner keeps sprinting down the alley and turns the corner.
Camera movement
The camera continues its slow push-in toward the subject.
Scene development
The car drives past the camera and disappears into the fog.
Timing or mood
Hold the same quiet, cinematic mood as the character looks out over the city.
Audio direction
Keep the city ambience and add a subtle rise in wind as the shot continues.
You can leave the prompt empty when you simply want the model to continue the scene naturally. Add a prompt when the clip has a clear next beat.
***
### Prompt Examples
#### Continue A Walk Cycle
{% code overflow="wrap" %}
```
The character keeps walking at the same pace, passing under the streetlights while the camera follows smoothly from behind.
```
{% endcode %}
Why it works:
* It continues the existing action.
* It preserves the camera relationship.
* It gives the model a clear next 7 seconds.
#### Finish A Reveal
{% code overflow="wrap" %}
```
The camera continues panning right to reveal the full mountain valley, holding the same golden-hour lighting and calm cinematic mood.
```
{% endcode %}
Why it works:
* It tells the camera what to do next.
* It keeps the same lighting and mood.
* It turns an unfinished move into a usable reveal.
#### Extend An Establishing Shot
```
The drone continues gliding forward over the coastline, waves crashing below, maintaining the same smooth cinematic movement.
```
Why it works:
* It asks for continuation, not reinvention.
* It reinforces motion, setting, and style.
#### Add Room Before A Cut
{% code overflow="wrap" %}
```
The subject holds their pose for a moment longer as the background movement continues naturally, leaving a clean ending for an edit point.
```
{% endcode %}
Why it works:
* It gives editorial intent.
* It asks for a usable tail, not more chaos.
***
### Best Practices
#### Start With A Strong Ending
The model extends from the end of the clip. If the last frame is blurry, chaotic, or mid-glitch, the continuation may inherit that problem. Trim to the strongest ending before extending.
#### Keep The Request Continuous
Use words like `continues`, `keeps`, `maintains`, `holds`, `glides`, or `moves forward`. These help frame the prompt as continuation.
#### Avoid Overloading The Next 7 Seconds
Do not ask for several new events, a major scene change, a character transformation, and a camera move all at once. One clean continuation usually works better.
#### Chain Extensions Carefully
You can extend more than once by using the extended result as the next input, but the total output cannot exceed 30 seconds. Each generation can also drift farther from the original, so review each extension before continuing.
#### Upscale Last
Because Extend outputs 720p, use it during the creative phase, then upscale only the version you plan to keep.
***
### Common Workflows
#### Quick Natural Extension
1. Attach a video that is 23 seconds or shorter.
2. Choose **Extend Generation**.
3. Leave the prompt empty.
4. Generate.
5. Review whether the continuation feels natural.
Best for simple camera moves, atmosphere, landscapes, and clips that already have obvious momentum.
#### Directed Continuation
1. Attach a video.
2. Choose **Extend Generation**.
3. Write what should happen next.
4. Keep audio on if you want generated sound for the continuation.
5. Generate and review.
Best for characters, action, reveals, product motion, or shots where the next beat matters.
#### Extend Then Finish
1. Generate or import a clip.
2. Extend it by 7 seconds.
3. Edit or trim the best section in Premiere.
4. Use Studio Upscale if the final result needs a higher-resolution finish.
Best for turning a short AI result into a more usable edit asset.
***
### Troubleshooting
#### The Video Is Too Long
Generative Extend can only accept videos up to 23 seconds because the output is capped at 30 seconds total. Trim the clip first, then extend.
#### The Output Is 720p
That is expected. Extend outputs 720p even if the source video is 1080p. Use Upscale after the creative result is approved.
#### The Aspect Ratio Is Not Supported
Generative Extend supports 16:9 and 9:16. Crop or resize square, ultra-wide, or unusual formats before extending.
#### The Extension Feels Random
Add a continuation prompt. Be specific about the next action, camera movement, and mood. Avoid describing a completely new scene.
#### The Audio Does Not Blend
Try another generation with audio off, or add a short audio direction in the prompt. For example: `Keep the room tone soft and natural, with no music swell.`
#### The Continuation Drifts After Multiple Extends
Chained extensions can gradually move away from the original shot. If the second or third extension drifts, trim back to the best version or generate a new clip from a better source frame.
***
### Related Pages
* [Video Generation](/features/video-generation) - Create, animate, transition, reference, extend, and edit videos.
* [Text-to-Video](/features/video-generation/text-to-video) - Generate a new shot from a prompt.
* [Image-to-Video](/features/video-generation/image-to-video) - Animate a still image into motion.
* [Video Canvas Editor](/features/video-generation/video-canvas-editor) - Open the classic editor from an existing video result.
* [Studio Upscale](/features/studio/video-upscaling) - Increase resolution after the extension is approved.
***
**Next:** If the extended clip is creatively right but too small, finish it with Studio Upscale.
# Image Generation
Chat Video Pro offers comprehensive AI image generation capabilities, from creating images from text to editing existing images, removing backgrounds, and upscaling resolution.
Image Generation is where you create and edit still images in Chat Video Pro. Use it for prompt-based images, image edits, thumbnails, transparent cutouts, image upscaling, and quick visual ideas.
If the image is meant to become part of a Studio video workflow, start with Studio Cinematic Lab. Cinematic Lab is designed for cinematic stills, source frames, look development, thumbnails, key art, and frames you may later animate in Motion Director, Multi-Cam, AI Transitions, or Relight Scene.
***
### What Image Generation Is For
Use Image Generation when you want to:
* Create an image from a text prompt.
* Edit or transform an existing image.
* Build a thumbnail concept.
* Remove a background.
* Upscale a final image.
* Create graphics, backgrounds, mood boards, product visuals, or social assets.
* Generate a still that can later become a video source image.
Use Studio when you want a more guided production workflow, especially for cinematic frames or images that feed directly into other Studio tools.
***
### Available Features
[**Supported Image Models**](/features/image-generation/supported-image-models)
Choose the right image model without reading a spec sheet. Use this page when you are deciding between Nano Banana, GPT Image 2, Flux, Seedream 5.0 Pro, Grok, Ideogram V4 Fast, Canvas Editor, background removal, or upscaling.
[**Text-to-Image**](/features/image-generation/text-to-image)
Create a new image from a written prompt. Use this for concepts, product visuals, backgrounds, thumbnail ideas, mood boards, and still assets.
[**Image-to-Image**](/features/image-generation/image-to-image)
Edit an existing image with a prompt. Use this when you already have a source image and want to change style, add or remove elements, adjust a visual direction, or create variations.
[**Canvas Editor**](/features/image-generation/canvas-editor)
Use a more controlled image editing workspace with layers, masks, annotations, and visual editing tools. Use this when a plain prompt is not enough and you need to guide exactly where changes should happen.
[**Background Removal**](/features/image-generation/background-removal)
Remove backgrounds from images and create transparent cutouts. Use this for products, portraits, thumbnails, graphics, and compositing.
[**Image Upscaling**](/features/image-generation/image-upscaling)
Increase image resolution after the image is approved. Use upscaling as a finishing pass, not as a way to fix a weak image.
[**Thumbnail Mode**](/features/image-generation/thumbnail-mode)
Create thumbnail-focused images with settings and workflows designed for clickable YouTube and social visuals.
***
### Image Generation vs. Studio Cinematic Lab
Both can create still images, but they are designed for different jobs.
Use Image Generation when...
Use Studio Cinematic Lab when...
You need a quick image or edit.
You need a cinematic frame or key art.
You want direct model control.
You want camera, lens, focal length, aperture, references, and look controls.
You are making a general graphic, background, or image asset.
The image may become a Motion Director, Multi-Cam, AI Transition, or Relight source.
You are editing an existing image.
You are building production-style source frames from scratch.
The practical rule: use Image Generation for normal image work, and use Cinematic Lab when the still needs to feel like part of a production.
***
### Getting Started
#### Quick Path
Use the top-right model selector and type a natural image request in chat.
Best for:
* Fast ideas.
* Casual images.
* Visual brainstorming.
* Moments where aspect ratio and quality settings are not critical.
#### Full Control
Enable **Generate Media** when you need control over model, aspect ratio, quality, resolution, or attached source images.
Best for:
* Production images.
* Specific output formats.
* Model comparisons.
* Image edits.
* Thumbnail or social deliverables.
#### Studio Path
Open **Studio** and choose **Cinematic Lab** when you want a cinematic still with production-style controls.
Best for:
* Hero frames.
* Key art.
* Thumbnail bases.
* Look development.
* Source frames for Motion Director, Multi-Cam, AI Transitions, or Relight Scene.
***
### Quick Reference
Goal
Best starting point
Quick image from chat
Top-right image model selector
Prompt-based image with settings
Text-to-Image
Edit an existing image
Image-to-Image
Controlled mask/layer edit
Canvas Editor
Transparent cutout
Background Removal
Higher-resolution final
Image Upscaling
YouTube thumbnail concept
Thumbnail Mode
Cinematic still for video workflow
Studio Cinematic Lab
Animate a still
Image-to-Video or Studio Motion Director
***
### Workflow Tips
#### Pick The Final Shape Early
Choose the aspect ratio for the destination: 16:9 for YouTube thumbnails and video frames, 9:16 for vertical social, 1:1 or 4:5 for feed posts, and wider ratios for banners or cinematic frames.
#### Generate Before You Upscale
Upscale only after the image is approved. If composition, text, lighting, face quality, or product shape is wrong, fix that first.
#### Use GPT Image 2 When Text Matters
If the image contains signs, labels, title cards, packaging, or readable words, try GPT Image 2 early. For the final thumbnail text, manual text placement may still be cleaner.
#### Use References For Consistency
When a person, product, location, or brand style needs to stay consistent, attach references or use Cinematic Lab references.
#### Send Strong Images to the Studio
A good still can become the start of a larger workflow:
* Animate it in Motion Director.
* Generate alternate angles in Multi-Cam.
* Bridge it to another frame in AI Transitions.
* Relight it in Relight Scene.
***
### Next Steps
* Choose a model with [Supported Image Models](/features/image-generation/supported-image-models).
* Create a still with [Text-to-Image](/features/image-generation/text-to-image).
* Edit an image with[ Image-to-Image](/features/image-generation/image-to-image).
* Build a cinematic frame with Studio [Cinematic Lab](/features/studio/cinematic-lab).
* Animate a finished still with [Image-to-Video](/features/video-generation/image-to-video) or Studio [Motion Director](/features/studio/motion-director).
# Supported Image Models
Chat Video Pro supports multiple state-of-the-art AI image generation models, each optimized for different use cases, quality levels, and workflows.
Chat Video Pro includes several image models because image generation has different jobs: cinematic frames, fast drafts, product visuals, thumbnails, text-heavy designs, reference-guided edits, background removal, and upscaling.
You do not need to memorize every model. Start with what you are trying to make, then choose the model or workflow that fits the job.
{% hint style="info" %}
If you are creating a cinematic frame that may become a video, start with Studio Cinematic Lab. It wraps image generation in camera, lens, focal length, aperture, reference, aspect ratio, and model controls so you can think like a DP instead of managing raw model settings.
{% endhint %}
***
### Quick Recommendations
If you need...
Start with...
Why
A cinematic still or source frame for video
Studio Cinematic Lab
Best guided workflow for photoreal frames, references, and video-ready stills.
A fast, high-quality general image
Nano Banana 2
Strong default for most image generation and reference-guided work.
The hardest version of a Nano Banana prompt
Nano Banana Pro
Better for complex reasoning, difficult references, or harder real-world details.
Text inside the image
GPT Image 2 or Ideogram V4 Fast
GPT Image 2 for signs and packaging; Ideogram for posters, UI mockups, and design layouts at roughly one second per generation.
Highest realism or final production stills
Flux 2 Max
Strong detail, realism, and polished image quality.
Typography, posters, UI mockups, or region-precise edits
Seedream 5.0 Pro
ByteDance Pro tier — stronger text and layouts; edit one element without wrecking the rest of the frame.
Fast social/mobile image drafts
Grok
Fast image generation with unusual mobile and panoramic formats.
Quick design drafts with readable text
Ideogram V4 Fast
Replaces Z-Image Turbo as the fast image model; matching edit path when you attach a reference.
Complex image editing with layers/masks
Canvas Editor
Better than plain prompting for controlled edits.
Transparent cutouts
Background Removal
Use the dedicated background-removal workflow.
Larger/sharper final image
Image Upscaling
Use a dedicated upscaler after the image is approved.
***
### The Simple Rule
Choose the model based on the hardest part of the image:
| Hardest part of the image | What to prioritize |
| -------------------------------------------- | -------------------------------------------------------- |
| It must look like a real production frame | Cinematic Lab, Nano Banana, or Flux. |
| It contains readable text | GPT Image 2 or Ideogram V4 Fast. |
| It needs a specific person, product, or look | Reference-friendly models or Cinematic Lab references. |
| It is a quick draft | Nano Banana 2, Grok, or Ideogram V4 Fast. |
| It needs precise editing | Seedream 5.0 Pro Edit, Canvas Editor, or image-to-image. |
| It needs a transparent background | Background Removal. |
| It is already good but too small | Image Upscaling. |
The best model is not always the most expensive model. The best model is the one that solves the specific visual problem.
***
### Model Guide
#### Nano Banana 2
Nano Banana 2 is the best default for most image generation in Chat Video Pro.
Use it for:
* Cinematic images.
* Character or product concepts.
* Reference-guided image generation.
* Fast visual exploration.
* Source frames for Motion Director, Multi-Cam, AI Transitions, or Relight Scene.
* Drafting multiple directions before choosing a final look.
Choose Nano Banana 2 when you want a strong first answer quickly.
#### Nano Banana Pro
Nano Banana Pro is the version to try when the prompt is harder.
Use it for:
* Complex multi-subject scenes.
* Rare locations or real-world references.
* Images where the relationship between objects must make sense.
* Prompts that need more careful reasoning.
* Final candidates where Nano Banana 2 is close but not quite enough.
Do not use Pro for everything by default. For normal drafts, Nano Banana 2 is usually faster and more efficient.
#### GPT Image 2
GPT Image 2 is the model to try when prompt adherence, composition, or text matters.
Use it for:
* Signs, posters, book covers, labels, packaging, UI, or readable words.
* Complex compositions.
* Multi-image editing and Canvas Editor workflows.
* Weird or specific prompts where logic matters.
* Images that need careful placement of several elements.
If an image contains text that needs to be readable, switch to GPT Image 2 early.
#### Flux 2 Max
Flux 2 Max is a strong choice for high-quality, realistic, final-looking images.
Use it for:
* Final production stills.
* High-detail images.
* Photoreal texture.
* Product visuals.
* Polished image outputs where speed is less important.
Flux is a good comparison pass when a Nano Banana result is good but you want to test a more premium finish.
#### Seedream 5.0 Pro
Seedream upgrades from Lite to **5.0 Pro** for text-to-image and edit. Use it when typography, posters, UI mockups, or dense layouts matter — or when you need a region-precise edit that changes one element without wrecking the rest of the frame.
Use it for:
* Posters, packaging, and multi-language text layouts.
* UI mockups and design comps.
* Creative concepts across many aspect ratios.
* Image edits that should keep the surrounding frame intact.
* 2K native output, with optional 4K via Crisp upscale.
Attach a reference image to switch into **Seedream 5.0 Pro Edit**.
#### Grok
Grok is useful for fast image generation and mobile-first formats.
Use it for:
* Social content.
* Quick drafts.
* Mobile aspect ratios.
* Panoramic or unusual formats.
* Simple image-to-image edits.
Use Grok when speed and format flexibility matter more than maximum polish.
#### Ideogram V4 Fast
Ideogram V4 Fast replaces Z-Image Turbo as the fast image model. It is built for text-in-image and design output at roughly one second per generation, with a matching edit path when you attach a reference.
Use it for:
* Quick posters, logos, and layout tests.
* Readable text in drafts without waiting on a heavier model.
* Fast concept checks before a GPT Image 2 or Seedream Pro polish pass.
* Image-to-image restyles that need to keep text and layout fidelity.
Move to GPT Image 2, Seedream 5.0 Pro, Nano Banana, or Flux once the direction is worth polishing.
***
### Which Workflow Should I Use?
Model choice matters, but workflow choice comes first.
| You want to... | Use... |
| -------------------------------------- | ---------------------------------------- |
| Create a normal image from a prompt | Text-to-Image |
| Edit an existing image with a prompt | Image-to-Image |
| Build a cinematic still for video work | Studio Cinematic Lab |
| Make a controlled layered edit | Canvas Editor |
| Remove the background | Background Removal |
| Increase resolution | Image Upscaling |
| Create YouTube thumbnail concepts | Thumbnail Mode |
| Animate a still image | Image-to-Video or Studio Motion Director |
If you are making an image that will become video, think ahead. A clean, well-framed still is easier to animate, relight, transition, or use for Multi-Cam later.
***
### Image Generation vs. Cinematic Lab
Use normal Image Generation when you want direct model control or a quick one-off image.
Use Cinematic Lab when the image is part of a production workflow.
| Use normal Image Generation when... | Use Studio Cinematic Lab when... |
| --------------------------------------------- | ------------------------------------------------------------------------------------ |
| You need a quick image in chat. | You are building a cinematic frame. |
| You already know which model to use. | You want camera, lens, focal length, and aperture controls. |
| The image is a simple asset or draft. | The still may become a Motion Director, Multi-Cam, AI Transition, or Relight source. |
| You want direct prompt/model experimentation. | You want references, visual consistency, and production-style look control. |
Cinematic Lab is usually the better choice for hero frames, thumbnails, key art, style frames, and source images for Studio video workflows.
***
### Text, Logos, And Readable Words
AI image models do not all handle text equally.
Use GPT Image 2 when the image includes:
* A sign.
* A label.
* Packaging text.
* A book title.
* A poster.
* UI text.
* A thumbnail with readable words.
For text-heavy images, keep the wording short and clear. Long paragraphs inside generated images are still difficult. For final thumbnails or titles, you may get better control by generating the image background first, then adding text in Premiere, Photoshop, or another design tool.
***
### References And Image Inputs
References are useful when the output should match something: a person, product, location, wardrobe, lighting style, or art direction.
Good references are:
* Clear.
* High-resolution enough to read.
* Consistent with the desired result.
* Focused on the subject or style you want to preserve.
* Not overloaded with conflicting looks.
Use fewer references when the prompt is being ignored. Use stronger references when the model is drifting too far from the subject.
For projects with recurring characters, products, or locations, build a small reference set and reuse it. Consistency improves when the model sees the same visual anchors across generations.
***
### Speed vs. Quality
Use faster models when:
* You are brainstorming.
* You are testing composition.
* You need several visual directions.
* The image is not final.
Use higher-quality models when:
* The image is client-facing.
* The image will become a video source frame.
* Details, faces, products, or realism matter.
* You are close to the final look.
A practical workflow:
1. Draft quickly.
2. Pick the strongest direction.
3. Refine the prompt or references.
4. Generate a higher-quality final.
5. Upscale only after the image is approved.
***
### Practical Starting Points
#### If you are new
Start with:
* **Nano Banana 2** for most image generation.
* **GPT Image 2** when text or complex composition matters.
* **Cinematic Lab** for cinematic frames and video-ready stills.
* **Ideogram V4 Fast** for quick design drafts with readable text.
#### If you are making video source frames
Start with:
* Cinematic Lab for controlled camera/lens look.
* Nano Banana 2 for fast photoreal exploration.
* GPT Image 2 if the frame contains readable text.
* Flux 2 Max when you want to compare a more polished realism pass.
Then send the result into Motion Director, Multi-Cam, AI Transitions, or Relight Scene.
#### If you are making thumbnails
Start with:
* Thumbnail Mode for thumbnail-specific composition.
* GPT Image 2 when text or layout precision matters.
* Nano Banana or Flux for strong faces, scenes, or image backgrounds.
For final thumbnail text, consider adding the text manually after generation for maximum control.
#### If you are editing an existing image
Start with:
* Image-to-Image for prompt-based edits.
* Canvas Editor for controlled edits with masks, layers, or annotations.
* Background Removal for cutouts.
* Image Upscaling after the edit is approved.
***
### Troubleshooting
#### The image looks generic
Add more specific visual direction: subject details, setting, lighting, material, texture, mood, and composition. If it should feel like a real frame, try Cinematic Lab.
#### Text in the image is wrong
Use GPT Image 2 or add the final text manually in a design tool. Keep generated text short.
#### The model ignores my reference
Use a clearer reference, reduce conflicting references, or make sure the prompt does not contradict the image. For recurring subjects, reuse the same reference set across generations.
#### The model follows the reference too much
Use fewer references or write a stronger prompt describing what should change. If all references look similar, the model may treat them as strict instructions.
#### The image is too small or soft
Do not regenerate endlessly just for resolution. Once the image is approved, use Image Upscaling.
#### I am not sure which model to pick
Start with Nano Banana 2. If the problem is text, switch to GPT Image 2 or Ideogram V4 Fast. If the problem is posters, UI mockups, or region-precise edits, use Seedream 5.0 Pro. If the problem is final realism, compare Flux 2 Max. If the problem is speed, use Ideogram V4 Fast or Grok. If the problem is cinematic framing, use Cinematic Lab.
***
### Next Steps
* Use [Text-to-Image](/features/image-generation/text-to-image) for normal prompt-based image generation.
* Use [Image-to-Image](/features/image-generation/image-to-image) when editing an existing image.
* Use Studio [Cinematic Lab](/features/studio/cinematic-lab) for cinematic stills and video-ready frames.
* Use [Canvas Editor](/features/image-generation/canvas-editor) for controlled edits.
* Use [Image Upscaling](/features/image-generation/image-upscaling) after the image is approved.
# Text-to-Image
Create images from scratch using text descriptions. Simply describe what you want to see, and Chat Video Pro generates it using AI.
Text-to-Image creates a still image from a written prompt. Use it when you want a new image, concept, thumbnail idea, product visual, style frame, background, or reference asset and you do not already have a source image to edit.
Chat Video Pro gives you two normal Text-to-Image paths: a fast chat path for quick ideas, and a full-control Generate Media path when you need model, aspect ratio, resolution, and quality settings. For cinematic stills and video-ready source frames, use Studio Cinematic Lab.
***
### When To Use Text-to-Image
Use Text-to-Image when you want to create a still from scratch:
* A quick concept image.
* A product or brand visual.
* A thumbnail background.
* A mood board image.
* A social post image.
* A background plate.
* A first draft before image editing.
* A source image for later video generation.
Choose another workflow when:
| You want to... | Use instead |
| --------------------------------------- | ---------------------------------------- |
| Create a cinematic frame for video work | Studio Cinematic Lab |
| Edit an existing image | Image-to-Image |
| Make a controlled layered edit | Canvas Editor |
| Remove a background | Background Removal |
| Increase resolution | Image Upscaling |
| Build a YouTube thumbnail | Thumbnail Mode |
| Animate the image into video | Image-to-Video or Studio Motion Director |
***
### Quick Path vs. Full Control
#### Quick Path
Use the quick path when you want a fast image during a conversation.
1. Set your default image model in the top-right model selector.
2. Type a natural language request in chat.
3. Example: `Generate an image of a cozy cabin in the snow at night.`
4. Chat Video Pro generates the image using default settings.
Best for:
* Fast ideas.
* Casual requests.
* Conversation flow.
* Images where exact aspect ratio and resolution are not critical.
Limitations:
* Less control.
* Uses default settings.
* Not ideal for production-specific format requirements.
#### Full Control
Use Generate Media mode when the image needs specific settings.
1. Enable **Generate Media** in the composer.
2. Choose an image model.
3. Choose aspect ratio and resolution/quality.
4. Enter a more detailed prompt.
5. Generate and review.
Best for:
* Production work.
* Specific aspect ratios.
* Higher-quality outputs.
* Model comparisons.
* Images that may become thumbnails, key art, or video source frames.
***
### Text-to-Image vs. Cinematic Lab
Use normal Text-to-Image when you want direct model control or a quick image.
Use Cinematic Lab when the still needs to feel like a real production frame.
| Use Text-to-Image when... | Use Studio Cinematic Lab when... |
| -------------------------------------------------- | ------------------------------------------------------------------------------------ |
| You need a quick image or concept. | You need a cinematic still with camera/lens control. |
| You know which image model you want. | You want the workflow to guide the visual look. |
| The image is a draft or simple asset. | The image may become a Motion Director, Multi-Cam, AI Transition, or Relight source. |
| You want to manually write the whole image prompt. | You want camera body, lens, focal length, aperture, references, and model controls. |
The practical rule: use Text-to-Image for general image generation, and use Cinematic Lab when the still is part of a video or production workflow.
***
### What A Good Prompt Includes
A strong image prompt usually describes:
Prompt element
What to include
Subject
The main person, object, place, or idea.
Setting
Where the image takes place.
Composition
Close-up, wide shot, centered product, low angle, overhead, etc.
Lighting
Golden hour, studio lighting, neon, soft window light, moody shadows.
Useful structure:
{% code overflow="wrap" %}
```
[Subject] in [setting], [composition], [lighting], [style], [color palette], [specific details], [mood].
```
{% endcode %}
You do not need to include everything every time. Add the details that actually matter for the image.
***
### Prompt Examples
#### Cinematic Concept Image
{% code overflow="wrap" %}
```
A detective standing in a rain-soaked alley at night, trench coat dripping, neon signs reflected in puddles, low-angle composition, moody cinematic lighting, shallow depth of field, tense noir atmosphere.
```
{% endcode %}
Why it works:
* Clear subject and setting.
* Lighting and mood are specific.
* Composition gives the model a shot shape.
#### Product Visual
{% code overflow="wrap" %}
```
Premium product photo of a matte black wireless headphone case on a dark stone surface, soft studio key light, subtle rim light, shallow depth of field, clean luxury commercial style, detailed texture.
```
{% endcode %}
Why it works:
* Product, material, surface, lighting, and style are all clear.
* The image has a usable commercial direction.
#### Social Background
{% code overflow="wrap" %}
```
Vibrant abstract background for a vertical social post, electric blue and magenta gradient, soft grain texture, subtle light streaks, clean center area for text overlay, modern energetic style.
```
{% endcode %}
Why it works:
* It tells the model the final use.
* It reserves space for text.
* It avoids overloading the image with detail.
#### Thumbnail Concept
{% code overflow="wrap" %}
```
High-contrast YouTube thumbnail background, dramatic studio lighting, shocked creator silhouette on the left, glowing laptop screen on the right, bold red and yellow color palette, empty space at top for title text.
```
{% endcode %}
Why it works:
* It includes layout.
* It leaves room for text.
* It uses thumbnail-specific visual language.
***
### Weak Prompts To Avoid
Too vague:
```
A picture.
```
Better:
{% code overflow="wrap" %}
```
Photoreal image of a cozy mountain cabin at night, warm light glowing from windows, snow falling, pine trees around the cabin, cinematic winter atmosphere.
```
{% endcode %}
Missing composition:
```
Coffee shop.
```
Better:
{% code overflow="wrap" %}
```
Wide interior photo of a cozy coffee shop, wooden tables, plants near large windows, warm morning light, soft depth of field, inviting editorial lifestyle style.
```
{% endcode %}
No use case:
```
Abstract background.
```
Better:
{% code overflow="wrap" %}
```
Abstract 16:9 background for a tech presentation, dark navy gradient, subtle circuit-like light patterns, clean empty center space, premium modern look.
```
{% endcode %}
***
### Choosing A Model
Use the model based on the hardest part of the image.
Square social posts, profile-style images, balanced compositions.
4:5
Instagram feed portraits and social graphics.
21:9
Cinematic widescreen frames and banners.
4:3 or 3:4
Editorial, vintage, portrait, or alternate framing.
Pick the final deliverable shape before generating. Cropping after generation can cut off important subjects, text areas, or composition lines.
***
### Best Practices
#### Describe The Image You Need, Not Just The Topic
"A watch" gives the model a topic. "Premium product photo of a black watch on a dark reflective surface with rim light" gives it an image.
#### Include Composition Early
Composition controls whether the image is usable. Say:
* Close-up portrait.
* Wide establishing image.
* Centered product shot.
* Low-angle hero shot.
* Overhead flat lay.
* Empty space on the right for text.
#### Save Text For The Right Workflow
If the image needs readable words, use GPT Image 2 or add the final text manually after generation. For thumbnails and graphics, generating the background first and adding final text yourself often gives the cleanest result.
#### Use References For Consistency
If a character, product, or brand look must stay consistent, attach reference images in a workflow that supports them or use Cinematic Lab references.
#### Generate A Few Directions Before Polishing
Do not over-optimize the first result. Generate a few directions, pick what is working, then refine.
#### Upscale Last
Do not use upscaling to fix a bad image. Choose the image first, then use Image Upscaling if it needs more resolution.
***
### Example Workflows
#### Quick Concept Image
1. Use the quick path in chat.
2. Ask for a simple visual idea.
3. If the direction works, regenerate with full control or move into Cinematic Lab.
#### Production Still
1. Use Generate Media or Cinematic Lab.
2. Choose the final aspect ratio.
3. Write a detailed prompt with composition and lighting.
4. Generate several directions.
5. Use the best image as a source for Motion Director, Multi-Cam, or AI Transitions.
#### Thumbnail Background
1. Choose 16:9.
2. Prompt for strong contrast and empty title space.
3. Avoid asking the model to create final text unless using GPT Image 2.
4. Add final thumbnail text manually for control.
#### Product Visual
1. Describe the product, material, surface, and lighting.
2. Use a clean composition.
3. Keep the prompt focused on the product.
4. Use Image-to-Image or Canvas Editor for refinements.
***
### Troubleshooting
#### The image feels generic
Add specific composition, lighting, material, and mood. Avoid one-word prompts. If you want a production frame, try Cinematic Lab.
#### The image has bad text
Use GPT Image 2 or add the text manually after generation. Keep generated text short.
#### The image is the wrong shape
Set the aspect ratio before generating. If the final platform is vertical, generate vertical from the start.
#### The model ignored an important detail
Move the detail earlier in the prompt and remove competing instructions. If the detail is a person, product, logo, or brand look, use references.
#### The output looks too AI-generated
Add concrete physical details: material, texture, imperfect surfaces, realistic lighting, lens feel, and environment. Cinematic Lab can help if you want a more grounded production-frame look.
#### The result is close but not final
Use the right follow-up workflow:
| Problem | Better next step |
| ------------------------------ | --------------------------------- |
| Need to edit part of the image | Image-to-Image or Canvas Editor |
| Need a transparent cutout | Background Removal |
| Need higher resolution | Image Upscaling |
| Need motion | Image-to-Video or Motion Director |
| Need a different angle | Multi-Cam |
***
### Related Pages
* [Supported Image Models](/features/image-generation/supported-image-models) - Choose the right model.
* [Image-to-Image](/features/image-generation/image-to-image) - Edit an existing image.
* [Canvas Editor](/features/image-generation/canvas-editor) - Make controlled image edits.
* [Thumbnail Mode](/features/image-generation/thumbnail-mode) - Create thumbnail-focused images.
* [Cinematic Lab](/features/studio/cinematic-lab) - Generate cinematic stills for production and video workflows.
* [Image-to-Video](/features/video-generation/image-to-video) - Animate a still image.
***
**Next:** If the still needs to become a video shot, use Image-to-Video or Studio Motion Director.
# Image-to-Image
Edit and modify existing images by uploading an image and describing the changes you want. Perfect for style transfer, adding elements, changing colors, or transforming images.
Image-to-Image edits an existing image with a prompt. Use it when the image is already close, but you want to change the style, background, colors, mood, composition, objects, or visual direction.
This is different from Text-to-Image. Text-to-Image starts from nothing. Image-to-Image starts from a source image and asks the model to transform it.
{% hint style="info" %}
**Think of Image-to-Image as prompt-guided revision.** The source image gives the model the starting point. Your prompt tells it what to change and what to preserve.
{% endhint %}
***
### What This Tool Is For
Use Image-to-Image when you want to:
* Restyle an image.
* Change the lighting or mood.
* Replace or adjust a background.
* Add a visual element.
* Remove a simple distraction.
* Create a variation of an existing design.
* Turn a rough concept into a more polished image.
* Prepare a still for later video generation.
Choose a more specific workflow when the job needs stronger control:
| You want to... | Use instead |
| ------------------------------------ | ---------------------------------------- |
| Build a cinematic still from scratch | Studio Cinematic Lab |
| Make a controlled mask/layer edit | Canvas Editor |
| Remove the entire background | Background Removal |
| Increase resolution | Image Upscaling |
| Create a new image without a source | Text-to-Image |
| Animate the image into video | Image-to-Video or Studio Motion Director |
***
### How To Use It
1. Enable[ **Generate Media**](/getting-started/interface-overview/generate-media-button) in the composer.
2. Attach the image you want to edit.
3. Choose an image model, or let Chat Video Pro switch to the matching edit version.
4. Describe what should change.
5. Choose aspect ratio, resolution, quality, or other available settings.
6. Generate and compare the result with the source image.
When you attach an image, Chat Video Pro can route supported image models into their edit versions. For example, a text-to-image model may switch to its Image-to-Image version once the source image is attached.
***
### When Image-to-Image Works Best
Image-to-Image is strongest when the edit is clear and focused.
Good fit
Example
Style change
Make this look like a cinematic 35mm film still with warm grain.
Lighting change
Make the scene feel like soft golden hour, preserving the same subject and pose.
Background change
Replace the plain studio background with a dark modern office.
Object addition
Add a small vintage camera on the table, keeping the rest of the scene unchanged.
Product variation
Change the bottle color to matte black while keeping the label readable.
Mood variation
Make this more premium and dramatic, with deeper shadows and subtle rim light.
It is weaker when the prompt asks for too many unrelated changes at once. If you need precise placement, masking, layered composition, or several separate edits, use Canvas Editor.
***
### Choosing A Model
You do not need to memorize every edit model. Start with the type of edit.
If the edit needs...
Try...
A strong default for most image edits
Nano Banana 2
Text, labels, signs, packaging, or complex composition
GPT Image 2 Edit or Ideogram V4 Edit
High realism or polished final quality
Flux 2 Max Edit
Region-precise edits that keep the rest of the frame
Seedream 5.0 Pro Edit
Fast creative variations
Grok Edit or Ideogram V4 Edit
More exact placement, masks, or layered control
Canvas Editor
Use Supported Image Models for the broader chooser guide.
***
### Writing Better Edit Prompts
A good edit prompt has three parts:
Part
What to say
Change
What should be different.
Preserve
What should stay the same.
Direction
The style, mood, lighting, or format you want.
Useful structure:
{% code overflow="wrap" %}
```
Change [specific element] to [new result], while preserving [important parts of the original image], with [style/mood/lighting].
```
{% endcode %}
You do not need this exact wording every time, but it helps avoid accidental changes.
***
### Prompt Examples
#### Style Transfer
{% code overflow="wrap" %}
```
Make this look like a cinematic 35mm film photograph with warm tones, soft halation, subtle grain, and natural contrast. Preserve the subject, pose, and composition.
```
{% endcode %}
Why it works:
* It names the style.
* It tells the model what not to change.
* It avoids asking for a new scene.
#### Background Replacement
{% code overflow="wrap" %}
```
Replace the background with a clean modern kitchen at morning light. Keep the person, clothing, pose, and camera angle the same.
```
{% endcode %}
Why it works:
* It changes only the background.
* It protects the subject identity and pose.
* It gives the replacement a clear visual direction.
#### Product Variation
{% code overflow="wrap" %}
```
Change the product color to matte forest green, keep the shape, logo placement, label text, lighting, and tabletop composition unchanged.
```
{% endcode %}
Why it works:
* It is specific about the product change.
* It protects brand and composition details.
* It is useful for testing design variants.
#### Social Version
{% code overflow="wrap" %}
```
Turn this into a vertical social image with more negative space at the top for text, brighter contrast, and a clean premium look. Preserve the main subject.
```
{% endcode %}
Why it works:
* It explains the destination format.
* It asks for room for text.
* It protects the subject.
#### Video-Ready Still
{% code overflow="wrap" %}
```
Make this frame more cinematic and video-ready: add soft directional window light from camera left, deeper background shadows, realistic texture, and a clean 16:9 composition. Preserve the subject identity and framing.
```
{% endcode %}
Why it works:
* It improves the source frame for later animation.
* It specifies light direction and composition.
* It avoids changing the core image.
***
### What To Preserve
If something matters, say so. Models may reinterpret parts of the image unless you protect them.
Common things to preserve:
* Subject identity.
* Face, pose, body shape, or wardrobe.
* Product shape, logo, label, or color.
* Camera angle and composition.
* Background layout.
* Text or UI elements.
* Lighting direction.
* Aspect ratio.
Example:
{% code overflow="wrap" %}
```
Change the background to a snowy forest, but preserve the subject's face, jacket, pose, camera angle, and 16:9 framing.
```
{% endcode %}
***
### Aspect Ratio And Framing
For simple edits, keep the same aspect ratio as the source image. This usually preserves composition and avoids unexpected cropping.
Change aspect ratio when you are intentionally reformatting:
Destination
Common ratio
YouTube or video frame
16:9
Vertical short or Reel
9:16
Feed post
1:1 or 4:5
Cinematic banner
21:9
When changing aspect ratio, tell the model how to reframe the image. For example: `Convert this to 9:16 with the subject centered and extra space above the head for title text.`
***
### Image-to-Image vs. Canvas Editor
Use Image-to-Image when a prompt can describe the edit clearly.
Use Canvas Editor when the edit depends on where something happens.
Use Image-to-Image when...
Use Canvas Editor when...
The whole image needs a style or mood change.
You need to paint, mask, or mark a specific area.
You want a quick variation.
You need controlled placement.
The edit is easy to explain in one prompt.
The edit has multiple layers or regions.
You are exploring directions.
You already know exactly what should change.
If the model changes the wrong part of the image twice, switch to Canvas Editor.
***
### Best Practices
#### Make One Main Change At A Time
Image-to-Image works better when each generation has a clear job. If you need a new background, new outfit, new lighting, and new format, do them in stages.
#### Use Strong Source Images
The source image sets the ceiling. A blurry face, broken hand, bad product shape, or messy composition can carry into the edit.
#### Tell The Model What To Keep
Do not only say what to change. Add what should remain stable.
#### Save Good Variations
If an edit is close, save it before trying another direction. A later edit can drift away from the best version.
#### Prepare Video Source Frames Carefully
If the edited image will become video, keep the composition clean, avoid tiny text, leave room for motion, and use the same aspect ratio as the final video when possible.
***
### Troubleshooting
#### The Edit Is Too Subtle
Use stronger language and make the change more specific. Try `make the sky a dramatic stormy sunset` instead of `make it nicer`.
#### The Image Changed Too Much
Add preservation language: `Keep the same subject, pose, camera angle, and composition.` You can also make a smaller edit first.
#### The Wrong Object Changed
Use location words like `foreground`, `background`, `left side`, `on the table`, or `behind the subject`. If the edit still lands in the wrong place, use Canvas Editor.
#### The Text Became Wrong
Use GPT Image 2 Edit when text matters, and keep text short. For final thumbnails or title graphics, it may be better to add text manually after generation.
#### The Aspect Ratio Cropped Important Details
Return to the source ratio or prompt the reframing directly: `keep the full body visible`, `leave space above the head`, or `do not crop the product`.
***
### Related Pages
* [Image Generation](/features/image-generation) - Create and edit still images.
* [Text-to-Image](/features/image-generation/text-to-image) - Generate a new image from a prompt.
* [Supported Image Models](/features/image-generation/supported-image-models) - Choose the right image model.
* [Canvas Editor](/features/image-generation/canvas-editor) - Use masks, layers, and visual edit controls.
* [Background Removal](/features/image-generation/background-removal) - Create transparent cutouts.
* [Studio Cinematic Lab](/features/studio/cinematic-lab) - Create cinematic stills and video-ready frames.
* [Studio Motion Director](/features/studio/motion-director) - Animate a finished still.
***
**Next:** If the edit needs exact placement or masking, open Canvas Editor. If the image is meant to become a cinematic video frame, use Studio Cinematic Lab.
# Canvas Editor
Advanced in-app image editing with layers, masks, annotations, and composition tools. Create complex edits without leaving Chat Video Pro.
Canvas Editor is the controlled image-editing workspace inside Chat Video Pro. Use it when you need to show the app where something should happen, not just describe the change in a prompt.
If Image-to-Image is prompt-guided revision, Canvas Editor is visual-guided revision. You still write a prompt, but you also give the model masks, boxes, arrows, text, or added images so it understands the exact area, placement, or composition you want.
{% hint style="info" %}
**Use Canvas when location matters.** If you have tried a normal Image-to-Image prompt twice and the model keeps changing the wrong part of the image, move into Canvas Editor.
{% endhint %}
***
### What This Tool Is For
Use Canvas Editor when you want to:
* Change one specific region of an image.
* Remove or replace a local object.
* Add a new element in a specific position.
* Point to what should move, change, or stay.
* Compose multiple images together.
* Place text, labels, badges, or layout notes.
* Give the model visual markup instead of relying only on language.
Use a different workflow when:
You want to...
Use instead
Make a global style or mood change
Image-to-Image
Generate a new image from scratch
Text-to-Image
Remove the entire background from an image
Background Removal
Build a cinematic still for video
Studio Cinematic Lab
Upscale a finished image
Image Upscaling
***
### Canvas Editor vs. Image-to-Image
Both edit existing images. The difference is how much control you need.
Use Image-to-Image when...
Use Canvas Editor when...
The whole image needs a style, mood, color, or background change.
The edit depends on a specific area.
The edit is easy to describe in one prompt.
You need to draw, mask, or point.
Speed matters more than exact placement.
Exact placement matters more than speed.
You want broad creative variations.
You want a directed revision.
Example:
* Image-to-Image prompt: `Make this scene feel like golden hour.`
* Canvas Editor prompt: `Remove this sign and rebuild the wall behind it.` with the sign brushed or boxed.
***
### How To Open Canvas Editor
#### From A Generated Image
1. Generate an image in Chat Video Pro.
2. Click **Edit** underneath the generation
3. Canvas Editor opens with the generated image loaded.
#### From An Uploaded Image
1. Upload or attach an image.
2. Click **Edit** on the image thumbnail.
3. Canvas Editor opens with the uploaded image ready to mark up.
***
### How Canvas Edits Work
Canvas Editor uses your visual markup to decide what kind of edit you are asking for.
What you do in Canvas
What it tells the model
Brush or draw over an area
This is the region to change, remove, or rebuild.
Draw a box or ellipse
This is the target area or object.
Add an arrow
This is the direction, location, or object being referenced.
Add text
This text or label should be part of the edit or layout.
Add another image
Use this as a composited element or visual reference.
You do not need to choose a technical pipeline. The editor reads your markup and sends the right context with your prompt.
***
### Available Tools
Brush tool
* Draw freehand selections
* Paint areas to modify
* Adjustable brush size
* Perfect for organic shapes
Rectangle Tool
* Draw rectangular selections
* Precise box selection
* Good for defined areas
* Multiple selections supported
Ellipse Tool
* Draw circular/oval selections
* Perfect for round objects
* Precise circular areas
Arrow Tool
* Draw arrows to indicate direction
* Show movement or rotation
* Point to specific areas
* Useful for repositioning
Text Tool
* Add text overlays
* Type annotations
* Label elements
* Create text graphics
Add Image
* Upload additional images
* Compose multiple elements
* Layer images together
* Create complex scenes
***
### Common Workflows
#### Remove A Specific Object
1. Open Canvas Editor.
2. Brush or box the object.
3. Prompt: `Remove this object and rebuild the background naturally.`
4. Apply the edit.
5. Review the result.
Best for small distractions, signs, logos, props, or background objects. For full transparent cutouts, use Background Removal.
#### Change One Part Of An Image
1. Select the area you want to change.
2. Prompt the specific change.
3. Add preservation language for everything else.
Example:
{% code overflow="wrap" %}
```
Change only the selected jacket to deep navy blue. Keep the person's face, pose, background, and lighting unchanged.
```
{% endcode %}
#### Add An Object In A Specific Spot
1. Draw a box where the object should go.
2. Prompt the object, scale, and style.
3. Include what should stay unchanged.
Example:
{% code overflow="wrap" %}
```
Add a small vintage film camera inside the selected area on the table. Match the scene lighting and keep the rest of the image unchanged.
```
{% endcode %}
#### Compose With Another Image
1. Open Canvas Editor.
2. Click **Add Image**.
3. Upload the product, logo, subject, prop, or reference asset.
4. Position it where it should appear.
5. Prompt how it should blend into the scene.
Example:
{% code overflow="wrap" %}
```
Blend the added product into the scene as if it was photographed there, matching perspective, shadows, and studio lighting.
```
{% endcode %}
#### Build A Thumbnail Layout
1. Open a strong image or thumbnail background.
2. Add text or arrows to block the layout.
3. Prompt the intended design.
4. Keep text short and clear.
Canvas is useful for thumbnail planning because placement matters. For a more thumbnail-specific workflow, use Thumbnail Mode.
***
### Prompting In Canvas
Canvas prompts should refer to your visual markup.
Good phrases:
* `In the selected area...`
* `Remove the brushed object...`
* `Replace the boxed region with...`
* `Place the added image naturally into the scene...`
* `Follow the arrow and move the object slightly to the right...`
* `Keep everything outside the selected area unchanged...`
Useful structure:
```
Change [the marked area] to [new result], while preserving [everything else that matters].
```
The most common mistake is writing a broad prompt after making a precise selection. If you selected one area, keep the prompt local.
***
### Choosing A Model
Canvas Editor works with the current image edit models in Chat Video Pro. Choose based on the hardest part of the edit.
If the edit needs...
Try...
Text, labels, layout, or complex instruction following
GPT Image 2 or Ideogram V4 Fast
Strong general image editing
Nano Banana 2
Polished realism or premium finish
Flux 2 Max
Region-precise edits or poster/UI layouts
Seedream 5.0 Pro
Fast creative variations
Grok or Ideogram V4 Fast
If one model keeps misunderstanding the markup, try GPT Image 2 for complex composition or Nano Banana 2 for a strong general pass.
***
### Best Practices
#### Select Only What Needs To Change
Do not brush half the image if only one object needs editing. Smaller, cleaner selections usually produce cleaner results.
#### Say What Should Stay The Same
Canvas tells the model where to look, but the prompt should still protect important details: subject identity, pose, product shape, text, lighting, framing, and background.
#### Use One Clear Edit Per Pass
For complex edits, work in stages:
1. Remove or replace the object.
2. Adjust lighting or style.
3. Add text or layout details.
4. Upscale only after the image is approved.
#### Use The Right Markup
Use brush for organic shapes, boxes for hard-edged areas, arrows for direction, and added images for composition. The cleaner your visual instruction, the less the model has to guess.
#### Keep Source Quality High
Canvas can guide the edit, but it cannot fully fix a weak source image. Use clear images with enough resolution for the model to understand the subject.
***
### Troubleshooting
#### The Wrong Area Changed
Make the selection smaller and add location language in the prompt, such as `only the selected sign`, `the object inside the box`, or `the brushed area in the background`.
#### The Edit Ignored My Markup
Use a more direct prompt that refers to the markup: `Use the selected area as the only edit region.` If that fails, try a different model or simplify the selection.
#### The Image Changed Too Much
Add preservation language: `Keep everything outside the selected area unchanged.` Also avoid asking for broad style changes in the same pass as a local edit.
#### Added Images Do Not Blend Naturally
Prompt for integration: `match lighting, perspective, shadows, color temperature, and depth of field.` If the added asset has a transparent background, it often works better as a clean product, logo, sticker, or prop.
#### Text Is Not Perfect
AI-generated text can still drift. Use GPT Image 2 for text-heavy edits, keep wording short, and add final production text manually if exact typography matters.
***
### Related Pages
* [Image Generation](/features/image-generation) - Create and edit still images.
* [Image-to-Image](/features/image-generation/image-to-image) - Edit an image with a prompt.
* [Text-to-Image](/features/image-generation/text-to-image) - Generate a new image from scratch.
* [Background Removal](/features/image-generation/background-removal) - Remove an image background.
* [Thumbnail Mode ](/features/image-generation/thumbnail-mode)- Create thumbnail-focused images.
* [Studio Cinematic Lab](/features/studio/cinematic-lab) - Build cinematic stills and video-ready frames.
***
**Next:** Use Image-to-Image for broad prompt-based edits, or use Background Removal when the whole image needs a transparent cutout.
# Background Removal
Automatically remove backgrounds from images using AI. Get transparent PNGs or replace backgrounds with new scenes. Perfect for product photos, portraits, and graphics.
Background Removal creates a transparent cutout from a still image. Use it when you want to keep the main subject and remove everything behind it.
This is the image workflow for product cutouts, portrait graphics, thumbnail subjects, stickers, overlays, and design assets. If you need to remove the background from a moving video, use Studio Rotoscope instead.
{% hint style="info" %}
**Background Removal answers: "What should stay in this image?"** The result is usually a transparent PNG that can be placed over another background, graphic, frame, or Premiere composition.
{% endhint %}
***
### What This Tool Is For
Use Background Removal when you want to:
* Create a transparent product cutout.
* Isolate a person, face, object, logo, animal, car, prop, or graphic.
* Make a thumbnail subject easier to composite.
* Remove a plain or distracting background from a still image.
* Build stickers, overlays, lower-third graphics, or social assets.
* Prepare a subject for Canvas Editor, Image-to-Image, Photoshop, Premiere, or another design tool.
Use a different workflow when:
You want to...
Use instead
Remove a background from video
Studio Rotoscope
Remove one object while keeping the original background
Image-to-Image or Canvas Editor
Replace the background with a new scene
Image-to-Image or Canvas Editor after creating the cutout
Make a controlled local edit
Canvas Editor
Upscale the final cutout
Image Upscaling
***
### Two Background Removal Paths
Chat Video Pro can route background removal in two main ways.
#### Automatic Background Removal
Use this for normal requests like:
```
Remove the background.
```
```
Make this transparent.
```
```
Create a cutout PNG.
```
Best for:
* One clear subject.
* Product photos.
* Portraits.
* Simple or medium-complexity backgrounds.
* Fast transparent PNGs.
This path uses Bria background removal to detect the main subject and remove the background automatically.
#### Subject-Specific Isolation
Use this when the image has multiple subjects or the app needs to know exactly what to keep.
```
Keep only the red sneaker and make everything else transparent.
```
```
Isolate the person on the left.
```
```
Remove the background, keep only the blue car.
```
Best for:
* Multiple people or objects.
* Busy scenes.
* Product groups where only one item should remain.
* Images where the main subject is ambiguous.
This path uses SAM 3 subject isolation for still images. Your wording matters because the model uses the subject description to decide what stays.
***
### How To Use It
1. Attach or upload an image.
2. Type a background removal request.
3. Be specific if there are multiple possible subjects.
4. Generate the result.
5. Use the transparent PNG in your edit, thumbnail, design, or composite.
If you only say `remove background`, Chat Video Pro tries to keep the main subject. If that is not the subject you wanted, run it again with a clearer `keep only...` prompt.
***
### Image Background Removal vs. Studio Rotoscope
Background Removal and Rotoscope both create cutouts, but they work on different media.
Need
Use
Transparent PNG from a still image
Background Removal
Transparent moving subject from video
Studio Rotoscope
Product or portrait cutout
Background Removal
Presenter, dancer, product demo, or moving foreground subject
Studio Rotoscope
One static frame from Premiere
Frame Capture, then Background Removal
A full video clip with transparency
Studio Rotoscope
The practical rule: **Background Removal is for images. Rotoscope is for video.**
***
### Best Source Images
Background Removal works best when the subject is easy to understand.
Good source images usually have:
* A clear foreground subject.
* Enough contrast between subject and background.
* Good lighting.
* Clean edges.
* Enough resolution to see hair, product details, or fine shapes.
* Minimal motion blur.
Harder source images include:
* Hair against a similar-colored background.
* Transparent glass, smoke, mesh, lace, or reflective objects.
* Crowded scenes with multiple possible subjects.
* Low-resolution screenshots.
* Subjects partly hidden behind other objects.
For difficult images, use a subject-specific prompt.
***
### Prompt Examples
#### Product Cutout
```
Remove the background and keep only the product. Make everything else transparent.
```
Good for e-commerce images, product graphics, thumbnails, ads, and catalogs.
#### Portrait Cutout
```
Remove the background, keep the person, and preserve the hair edges as cleanly as possible.
```
Good for headshots, thumbnails, creator graphics, social posts, and presenter overlays.
#### Specific Subject In A Busy Scene
```
Keep only the woman in the red jacket and make everything else transparent.
```
Good when the image contains multiple people or objects.
#### Logo Or Graphic Isolation
```
Isolate the logo and make the rest of the image transparent.
```
Good for design assets, overlays, and branded graphics.
***
### Common Workflows
#### Quick Transparent Cutout
1. Attach the image.
2. Type: `Remove the background.`
3. Generate the transparent PNG.
4. Use it in your edit or design.
Best for clear portraits, products, or simple subjects.
#### Subject-Specific Cutout
1. Attach the image.
2. Describe exactly what should remain.
3. Generate the cutout.
4. Review the edges and subject choice.
Example:
```
Keep only the black camera on the table and make everything else transparent.
```
Best for busy images or multiple-subject scenes.
#### Cutout Then Composite
1. Remove the background from the subject image.
2. Open Canvas Editor or Image-to-Image.
3. Place the cutout over a new background or prompt a new scene.
4. Match lighting, shadows, and scale.
Best when you want a polished creative composite instead of a simple transparent asset.
#### Cutout For Thumbnail Design
1. Remove the background from the person, product, or object.
2. Place the cutout over a bold thumbnail background.
3. Add title text manually or with Thumbnail Mode.
4. Keep the silhouette large and readable.
Best for YouTube thumbnails, social graphics, and ad creatives.
***
### Best Practices
#### Be Specific When There Are Multiple Subjects
If the image has several people or products, do not only say `remove background`. Say `keep only the person on the left` or `keep only the red shoe`.
#### Use Cutouts As Building Blocks
A clean transparent subject can be reused across many designs: thumbnails, ads, lower thirds, social posts, title cards, and Premiere composites.
#### Do Not Use Background Removal For Object Erasing
If you want to remove an object and keep the original background, that is an image editing/inpainting task. Use Canvas Editor or Image-to-Image.
#### Check Edges Before Final Design
Look closely at hair, hands, product edges, transparent materials, and shadows. If the edge is rough, try a higher-quality source or a more specific subject prompt.
#### Upscale After The Cutout Is Approved
If the cutout is correct but too small, use Image Upscaling after the background removal result is approved.
***
### Troubleshooting
#### The Wrong Subject Was Kept
Use a subject-specific prompt:
```
Keep only the person wearing the yellow jacket and make everything else transparent.
```
#### The Background Was Not Fully Removed
Try a clearer prompt like `make the entire background transparent` or use a higher-quality image with better contrast.
#### The Edges Are Rough
Use a sharper source image, avoid heavy blur, and make sure the subject is well lit. Fine hair, glass, mesh, and shadows are naturally harder.
#### It Removed Part Of The Subject
Describe the subject more completely. For example: `keep the entire person including hair, hands, and shoes.`
#### I Need This For Video
Use Studio Rotoscope. Background Removal is for still images, while Rotoscope tracks a selected subject through a moving video clip.
***
### Related Pages
* [Image Generation](/features/image-generation) - Create and edit still images.
* [Image-to-Image](/features/image-generation/image-to-image) - Edit or restyle an image with a prompt.
* [Canvas Editor ](/features/image-generation/canvas-editor)- Use masks, markup, and added images for controlled edits.
* [Image Upscaling](/features/image-generation/image-upscaling) - Increase resolution after the cutout is approved.
* [Thumbnail Mode](/features/image-generation/thumbnail-mode) - Build thumbnail-focused images and layouts.
* [Studio Rotoscope](/features/studio/sam-3-rotoscoping) - Remove video backgrounds and create transparent moving subjects.
***
**Next:** Use Image Upscaling if the cutout is correct but needs more resolution, or Studio Rotoscope if the source is video.
# Image Upscaling
Increase image resolution using AI upscaling models. Enhance low-resolution images, upscale generated images, or prepare images for print.
{% embed url="" %}
{% hint style="info" %}
**Upscale last.** Fix composition, prompt accuracy, text, faces, product shape, background removal, and image edits first. Then upscale the version you actually want to keep.
{% endhint %}
***
### What This Tool Is For
Use Image Upscaling when you want to:
* Increase the resolution of a generated image.
* Make a final thumbnail background or subject cutout larger.
* Prepare an image for client review, web use, print, or a higher-resolution edit.
* Improve a low-resolution image that is otherwise useful.
* Create a cleaner final still after Image-to-Image, Canvas Editor, or Background Removal.
* Enhance a source frame before using it in a video workflow.
Use a different workflow when:
| You want to... | Use instead |
| ------------------------------------ | ------------------------------- |
| Upscale video | Studio Upscale |
| Fix composition, style, or content | Image-to-Image or Canvas Editor |
| Create a new image | Text-to-Image |
| Create a cinematic video-ready still | Studio Cinematic Lab |
| Remove a background before upscaling | Background Removal |
***
### How To Use It
1. Start with a generated or uploaded image.
2. Click **Transform** on the image thumbnail.
3. Choose an image upscaling model.
4. Choose the scale factor or model-specific settings.
5. Generate the upscale.
6. Review the result before using it in your final project.
You can also ask in natural language:
```
Upscale this image.
```
```
Make this higher resolution.
```
```
Enhance the quality of this image.
```
Chat Video Pro can detect the upscaling intent and route the image to an upscaling model.
***
### Image Upscaling vs. Studio Upscale
Image Upscaling and Studio Upscale solve the same kind of problem for different media.
Need
Use
Higher-resolution still image
Image Upscaling
Higher-resolution transparent PNG cutout
Image Upscaling
Higher-resolution thumbnail, product image, or key art
Image Upscaling
Higher-resolution generated or imported video
Studio Upscale
Higher-resolution clip after Add Effects, Reshoot, or Rotoscope
Studio Upscale
The practical rule: **Image Upscaling is for stills. Studio Upscale is for video.**
***
### Choosing An Image Upscaler
Start with the source image and the problem you are solving.
| If you need... | Try... |
| ----------------------------------------------- | --------------------- |
| Best general quality and control | Topaz Image Upscaler |
| Fast, crisp, simple enhancement | Recraft Crisp Upscale |
| Restoration for compressed or real-world images | Swin2SR Restore |
#### Topaz Image Upscaler
Topaz is the most flexible image upscaler.
Use it when:
* You want a reliable professional upscale.
* You need 1.5x, 2x, 3x, or 4x scaling.
* You are working with portraits, faces, product shots, or final stills.
* You want JPEG or PNG output.
* You want face enhancement.
Useful Topaz modes include:
Mode
Use when
Standard V2
You want the safest general-purpose upscale.
High Fidelity V2
The image is already good and you want to preserve detail.
Low Resolution V2
The source image is small or soft.
Text Refine
The image contains text or signage.
Recovery / Recovery V2
The image is compressed, damaged, or low quality.
CGI
The image is synthetic, graphic, or computer-generated.
#### Recraft Crisp Upscale
Recraft Crisp Upscale is a simple detail-focused option.
Use it when:
* You want a fast, clean upscale.
* You do not need many settings.
* The image is already good and just needs more crispness.
* You want a straightforward finishing pass after generation.
#### Swin2SR Restore
Swin2SR is useful for restoration-style tasks.
Use it when:
* The image is compressed.
* The source is a real-world photo that needs repair.
* You want to compare a restoration pass against Topaz.
* The image needs super-resolution rather than a purely crisp upscale.
Swin2SR exposes task modes like Classical SR, Compressed SR, and Real-World SR.
***
### Scale Factor Guide
Choose the smallest scale that solves the delivery problem.
Scale
Use when
1.5x
You need a small boost and want to avoid over-processing.
2x
You want the safest default upscale.
3x
You need a larger output but the source is already clean.
4x
You need maximum size or are preparing a very small image for larger use.
More scale is not always better. A 4x upscale can make artifacts, bad text, strange hands, or noisy edges more obvious. If the source image has problems, fix those before upscaling.
***
### Common Workflows
#### Final Generated Image Upscale
1. Generate or edit the image until the composition is approved.
2. Click **Transform**.
3. Choose Topaz or Recraft.
4. Use 2x for a normal finish, or 4x when you need a larger output.
5. Review the result at full size.
Best for final stills, social graphics, and image assets.
#### Thumbnail Finishing Pass
1. Create the thumbnail image or background.
2. Make sure faces, text areas, subject cutouts, and composition are right.
3. Upscale after the design direction is approved.
4. Add final title text manually if exact typography matters.
Best for YouTube thumbnails and social cover images.
#### Product Or Portrait Cutout Upscale
1. Use Background Removal to create the cutout.
2. Check the edges.
3. Upscale the approved cutout.
4. Place it into the final design or Premiere project.
Best for ads, product graphics, creator thumbnails, and overlays.
#### Restoration Pass
1. Upload the low-resolution or compressed image.
2. Try Topaz Recovery V2 or Swin2SR.
3. Compare outputs.
4. Use the cleaner result as the new source.
Best for old images, screenshots, compressed web images, and rough client-provided assets.
***
### Best Practices
#### Do Creative Work First
Upscaling should not be used to rescue a bad image. If the subject, lighting, text, product shape, or composition is wrong, fix those first.
#### Inspect The Result At Full Size
Zoom in and check faces, hands, text, logos, product edges, hair, and transparent cutout edges. Upscaling can improve detail, but it can also reveal problems.
#### Use PNG When Transparency Or Quality Matters
Use PNG for transparent cutouts, design assets, and files you plan to composite further. Use JPEG when file size and compatibility matter more.
#### Avoid Repeated Upscaling
Do not upscale the same image again and again unless you are intentionally testing. Repeated upscales can create artificial texture and artifacts.
#### Keep A Copy Of The Pre-Upscale Image
Save the approved source before upscaling. If the upscaled version introduces artifacts, you can rerun with a different model or scale.
***
### Troubleshooting
#### The Image Looks Sharper But Not Better
Upscaling adds resolution, but it does not automatically fix weak composition, bad anatomy, poor text, or an unclear subject. Return to Image-to-Image, Canvas Editor, or Text-to-Image first.
#### Faces Look Strange
Try Topaz with face enhancement, use a lower scale factor, or improve the source portrait before upscaling.
#### Text Still Looks Wrong
Use GPT Image 2 or Canvas Editor before upscaling if the text is part of the image. For final thumbnails or graphics, add exact text manually after the image is upscaled.
#### Edges Look Rough On A Transparent Cutout
Check the cutout before upscaling. If the alpha edge is bad, rerun Background Removal or clean the cutout before increasing resolution.
#### Upscaling Takes Too Long
Use a smaller scale, try Recraft Crisp Upscale, or upscale only the final selected image instead of every draft.
#### I Need To Upscale A Video
Use Studio Upscale. Image Upscaling only works on still images.
***
### Related Pages
* [Image Generation](/features/image-generation) - Create and edit still images.
* [Supported Image Models](/features/image-generation/supported-image-models) - Choose an image model before upscaling.
* [Text-to-Image](/features/image-generation/text-to-image) - Generate a new still.
* [Image-to-Image](/features/image-generation/image-to-image) - Fix or edit an image before upscaling.
* [Background Removal ](/features/image-generation/background-removal)- Create transparent image cutouts.
* [Thumbnail Mode](/features/image-generation/thumbnail-mode) - Build thumbnail-focused images.
* [Studio Upscale](/features/image-generation/image-upscaling) - Upscale video clips.
***
**Next:** Use Thumbnail Mode for thumbnail-focused images, or Studio Upscale when the source is video.
# Thumbnail Mode
Optimized settings and workflows for creating YouTube thumbnails. Includes aspect ratio presets, text overlay guidance, and best practices for click-worthy thumbnails.
{% embed url="" %}
Thumbnail Mode optimizes image generation for clickable YouTube and social thumbnails. Use it when the goal is not just "make an image," but "make a still that can earn attention in a feed."
***
### What Thumbnail Mode Is For
Use Thumbnail Mode when you want to:
* Generate YouTube thumbnail concepts.
* Explore multiple thumbnail directions for the same video.
* Create strong faces, subjects, backgrounds, and visual hooks.
* Build thumbnail backgrounds with room for title text.
* Use proven thumbnail patterns as inspiration.
* Create A/B test options before committing to one design.
Use a different workflow when:
You want to...
Use instead
Build cinematic key art or a premium hero frame
Studio Cinematic Lab
Generate text-heavy thumbnail designs
GPT Image 2 or Canvas Editor
Cut out a person or product for a thumbnail
Background Removal
Make precise layout edits
Canvas Editor
Upscale the final thumbnail
Image Upscaling
***
### Thumbnail Mode vs. Cinematic Lab
Both can help with thumbnails, but they solve different problems.
| Use Thumbnail Mode when... | Use Cinematic Lab when... |
| ------------------------------------------------------------ | ------------------------------------------------------------------------------------ |
| You want thumbnail-specific psychology and A/B variations. | You want a cinematic still, key art, or production frame. |
| The thumbnail needs bold layout, contrast, and click appeal. | The image needs camera, lens, lighting, and filmic realism. |
| You want multiple concepts quickly. | You want one strong hero frame to build around. |
| The thumbnail is the final deliverable. | The frame may also become Motion Director, Multi-Cam, or AI Transition source media. |
A strong workflow is to make the hero image in Cinematic Lab, then use Canvas Editor or Thumbnail Mode thinking to adapt it into a thumbnail.
***
### How Thumbnail Mode Works
When Thumbnail Mode is enabled, Chat Video Pro adds a thumbnail strategy pass before image generation.
It can:
* Understand the topic, niche, and emotional angle.
* Apply thumbnail best practices.
* Improve the prompt for feed readability.
* Suggest stronger subject, layout, contrast, and hook direction.
* Generate multiple variations for A/B testing.
* Use blueprint references when Blueprint mode is enabled.
You still need to give it a clear idea. Thumbnail Mode improves a direction; it does not replace having a strong video promise.
***
### Thumbnail Mode Options
#### Off
Use Off for normal image generation or when editing an existing thumbnail.
Best for:
* Regular images.
* Image-to-Image edits.
* Canvas Editor refinements.
* Backgrounds or assets that are not thumbnails.
#### On
Use On when you want thumbnail optimization without visual blueprint references.
Best for:
* Quick thumbnail concepts.
* A clear idea that needs better execution.
* Faster generation.
* Testing hooks, poses, backgrounds, or visual metaphors.
#### On With Blueprints
Use On with Blueprints when you want the generation to use proven thumbnail layouts or style references as inspiration.
Best for:
* Professional thumbnail ideation.
* Finding a stronger layout direction.
* Matching a proven thumbnail style.
* Exploring formats you would not have prompted manually.
Blueprints should inspire the structure, not copy another creator's thumbnail exactly.
***
### Good Thumbnail Inputs
A strong thumbnail prompt usually includes:
Close-up face, split screen, object hero, before/after, empty text space.
Visual hook
What makes the thumbnail understandable in one second.
Useful structure:
{% code overflow="wrap" %}
```
Create a YouTube thumbnail for [video topic] that communicates [promise]. Show [subject/scene] with [emotion], high contrast, clear focal point, and space for short title text.
```
{% endcode %}
***
### Prompt Examples
#### Creator / Talking Head
{% code overflow="wrap" %}
```
Create a YouTube thumbnail for a video about fixing bad AI video results. Show a frustrated creator looking at a messy generated clip on a monitor, high contrast, expressive face, red warning accents, clean empty space on the left for title text.
```
{% endcode %}
Why it works:
* It names the video promise.
* It gives a human emotion.
* It leaves room for final text.
#### Product / Review
{% code overflow="wrap" %}
```
Create a YouTube thumbnail for a camera review comparing a cheap lens to an expensive lens. Show both lenses large in the foreground, dramatic split-lighting, clear versus composition, premium tech-review style, bold negative space for text.
```
{% endcode %}
Why it works:
* It sets up a comparison.
* It uses a simple visual structure.
* It gives the viewer a reason to be curious.
#### Tutorial / Outcome
{% code overflow="wrap" %}
```
Create a YouTube thumbnail for a Premiere Pro tutorial about cutting podcasts faster with AI. Show a timeline transforming from messy to clean, a confident editor at the desk, bright green success accents, clean professional YouTube tutorial style.
```
{% endcode %}
Why it works:
* It visualizes the result.
* It avoids a generic "person at computer" prompt.
* It gives the image a before/after idea.
#### Cinematic Thumbnail Background
{% code overflow="wrap" %}
```
Create a dramatic 16:9 thumbnail background for a video about relighting dull footage. Show a split scene with flat gray lighting on one side and cinematic golden rim light on the other, no text, strong contrast, clean center subject.
```
{% endcode %}
Why it works:
* It asks for a background, not final text.
* It creates a clear before/after visual hook.
* It is ready for manual title placement.
***
### Multi-Thumbnail Variations
When you request multiple thumbnails, Thumbnail Mode can create different approaches to the same idea.
Ask for variations when you are not sure which angle is strongest.
Examples:
```
Create 4 thumbnail variations for this video idea, each with a different visual hook.
```
{% code overflow="wrap" %}
```
Generate 3 A/B thumbnail options: one emotional face version, one object-focused version, and one before/after version.
```
{% endcode %}
Good variation types:
Story videos, reveals, experiments, unusual results.
Do not generate four nearly identical thumbnails. Ask for different hooks, not just different colors.
***
### Choosing A Model
Use models based on the hardest part of the thumbnail.
Need
Try
Strong all-around thumbnail concepts
Nano Banana 2 / Pro
Readable text, signs, UI, or layout precision
GPT Image 2
Fast early ideas
Nano Banana 2 or another fast supported image model
For final thumbnail text, manual typography is often still the best choice. Generate the visual, then add exact text in Canvas Editor, Premiere, Photoshop, or your design tool.
***
### Editing A Thumbnail
Drag the image back into the composer to continue editing.
#### Quick Prompt Edit
1. Re-attach or reuse the generated thumbnail.
2. Turn Thumbnail Mode **Off**.
3. Use Image-to-Image for a focused edit.
4. Prompt the exact change.
Example:
```
Make the background darker and keep the person's face, pose, and layout unchanged.
```
#### Precise Canvas Edit
1. Click **Edit** on the thumbnail.
2. Use Canvas Editor.
3. Mark the exact area to change.
4. Add or adjust text, objects, arrows, products, or cutouts.
For precision edits, keep Thumbnail Mode Off unless you intentionally want to generate a new optimized thumbnail direction.
#### Cutout Workflow
1. Use Background Removal to isolate a person or product.
2. Place the cutout on a bold thumbnail background.
3. Add title text and accents.
4. Upscale the final thumbnail if needed.
This is often better than asking one prompt to do everything.
***
### Best Practices
#### Make The Promise Visual
The thumbnail should show the reason to click. "Premiere Pro tutorial" is a topic. "Messy timeline becomes clean in one click" is a visual promise.
#### Use Fewer Elements
Small thumbnails punish clutter. One face, one product, one strong before/after, or one clear visual metaphor usually beats a busy scene.
#### Leave Room For Text
If you plan to add title text later, prompt for empty space. Say `clean empty space on the left for title text` or `simple background with room for two large words`.
#### Treat Text As A Finishing Step
AI can generate text, especially with GPT Image 2, but final thumbnail typography often looks best when added manually.
#### Generate Concepts Before Polishing
Use Thumbnail Mode for exploration. Pick the strongest concept, then refine it with Image-to-Image, Canvas Editor, Background Removal, and Upscaling.
#### Match The Hook To The Video
Do not make a thumbnail promise the video does not pay off. Strong thumbnails create clicks, but accurate thumbnails create trust.
***
### Common Workflows
#### Fast Thumbnail Concept
1. Enable Generate Media.
2. Choose a supported image model.
3. Turn Thumbnail Mode On.
4. Prompt the topic, promise, subject, and emotion.
5. Generate 2-4 options.
#### Cinematic Thumbnail
1. Use Studio Cinematic Lab for the hero frame.
2. Save the best still.
3. Use Canvas Editor to add layout, cutouts, or text space.
4. Add final text manually or with GPT Image 2.
5. Upscale when approved.
#### Thumbnail With Cutout Subject
1. Generate or upload a subject image.
2. Use Background Removal to create a transparent cutout.
3. Generate or design the background.
4. Compose in Canvas Editor.
5. Finish with Image Upscaling if needed.
***
### Troubleshooting
#### Thumbnail Mode Is Not Available
Use Generate Media and choose a supported image model. Availability can vary by model and release.
#### The Thumbnail Looks Generic
Add a clearer promise, emotion, and composition. Instead of `thumbnail for AI editing`, try `frustrated editor looking at a broken AI clip, red warning symbols, clear before/after layout`.
#### The Text Is Wrong
Use GPT Image 2 for text-heavy generations, keep generated text short, or add final text manually after generation.
#### Variations Look Too Similar
Ask for different concepts, not just multiple outputs. Specify variation types: reaction face, before/after, product hero, versus, mystery, or tutorial outcome.
#### The Result Is Too Busy
Reduce the number of subjects, simplify the background, and leave more negative space. Thumbnails must read quickly at small size.
#### I Need To Edit A Thumbnail
Turn Thumbnail Mode Off and use Image-to-Image or Canvas Editor. Thumbnail Mode is best for generating new thumbnail concepts.
***
### Related Pages
* [Image Generation ](/features/image-generation)- Create and edit still images.
* [Supported Image Models](/features/image-generation/supported-image-models) - Choose a model for thumbnail generation.
* [Text-to-Image](/features/image-generation/text-to-image) - Create normal prompt-based images.
* [Canvas Editor](/features/image-generation/canvas-editor) - Make precise thumbnail edits.
* [Background Removal ](/features/image-generation/background-removal)- Create cutout subjects.
* [Image Upscaling](/features/image-generation/image-upscaling) - Finish approved thumbnails at higher resolution.
* Studio Cinematic Lab - Create cinematic stills and hero frames.
* High-CTR Thumbnail Workflow - Full thumbnail production workflow.
***
**Next:** Use Canvas Editor for precise layout edits, or Image Upscaling when the final thumbnail needs more resolution.
# Studio
Studio is the fastest way to use Chat Video Pro when you already know the creative job you want done. Instead of writing a broad prompt in chat and hoping the right tool is selected, you open a purpose-built workflow: Cinematic Lab for master stills, Multi-Cam for alternate angles, Motion Director for camera moves, AI Transitions for shot bridges, Rotoscope for subject isolation, Relight Scene for lighting changes, and more.
{% hint style="info" %}
**Think of Studio as your production department inside Premiere.** Chat is still best for conversation, planning, assistant-style help, and flexible prompting. Studio is best when the job has a known shape and benefits from a guided interface, source media picker, presets, visual controls, and a dedicated results screen.
{% endhint %}
#### What Studio Is For
Studio is built around **workflows**. Each card opens a focused creative path with the right inputs, controls, and output surface already prepared.
Use Studio when you want to:
* **Create new shots** from scratch with cinematic camera and lens controls
* **Generate alternate camera angles** from a still or clip
* **Animate still images** with directed camera movement
* **Bridge two frames** into a seamless AI transition
* **Clean up footage** by removing backgrounds, erasing objects, or reshooting problem areas
* **Improve finished clips** with upscaling, motion capture, relighting, or effects passes
* **Move faster than prompt-only workflows** because the UI already knows what the job needs
Most Studio workflows end by sending the generated result back into your chat as a normal image or video message. From there, you can save it, open it, reuse it from Recents, drag it into Premiere, or feed it into another Studio workflow.
{% hint style="warning" %}
**Studio does not replace the main chat.** It gives you guided creative workflows for media generation and video editing. Use the chat when you need planning, troubleshooting, general Premiere help, or open-ended creative direction. Use Studio when you want to run a specific production task.
{% endhint %}
#### Opening Studio
1. Open Chat Video Pro inside Premiere Pro
2. Click the **Studio** button in the sidebar
3. Choose a card from the Launchpad
The Launchpad opens as a full-screen workspace over the main chat. It is organized like a production house: **Production**, **Post-Production**, and **Audio**.
***
#### Launchpad Navigation
The Launchpad is designed to stay out of the way once you know where everything lives.
* **Click a card** to open that workflow
* **Use Search** to find workflows by job, model, or keyword, such as `upscale`, `transition`, `rotoscope`, `camera move`, `remove object`, or `relight`
* **Use Back** in the Studio header to move from a workflow back to the Launchpad
* **Use Close** from the Launchpad when you want to leave Studio entirely
***
#### Card Badges and States
Studio cards can show a few different states:
| State | What It Means |
| --------------- | -------------------------------------------------------------------------------------------------------------------- |
| **NEW** | A newly launched workflow. These are ready to use, but they may still be expanding with presets, examples, and docs. |
| **Coming Soon** | A visible preview of a planned workflow. The card is locked and cannot be opened yet. |
As of this Studio release, the working workflows are concentrated in **Production** and **Post-Production**. The **Audio** department is visible on the Launchpad, but its cards are currently Coming Soon.
***
#### How Studio Workflows Usually Start
Most Studio workflows begin by asking for source media. The source picker is context-aware: it only shows the inputs a workflow can actually use.
For example:
* **Motion Director** asks for a single image because it animates stills
* **Rotoscope**, **Erase Objects**, **Add Effects**, **Reshoot**, and **Upscale** ask for video because they operate on clips
* **Multi-Cam** accepts either an image or a video because it can generate alternate still angles or run a video multicam move
* **Motion Capture** asks for two assets: a motion reference video and a character image
* **Cinematic Lab** skips the loader because it starts from a written scene description
* **AI Transitions** opens its own start-frame and end-frame picker inside the workflow
* **Relight Scene** opens its own image/video setup screen because it supports both still relighting and optional video relighting
***
#### Source Options
Depending on the workflow, the asset loader can pull media from:
| Source | Use It When |
| ----------------- | ------------------------------------------------------------------------ |
| **Upload** | The file is on your computer and not already in Chat Video Pro |
| **Recents** | You want to reuse something you generated, captured, or imported earlier |
| **Frame Capture** | You want a still from the current Premiere Pro playhead position |
| **Clip Import** | You want to send a selected Premiere clip into a video workflow |
The loader filters by workflow. If a workflow only supports images, you will not be asked to import video. If it only supports video, the picker focuses on clips. If it needs two inputs, such as Motion Capture, each slot explains what it expects.
{% hint style="info" %}
**Pro tip:** Treat Recents like your internal production shelf. Generate a strong Cinematic Lab still, send it to chat, then reuse it from Recents in Motion Director, Multi-Cam, AI Transitions, or Relight Scene. Studio is fastest when you chain outputs instead of hunting files on disk.
{% endhint %}
#### Production Department
Production workflows create new shots, new angles, and new motion. Start here when you are building media that did not already exist in your edit.
| Workflow | Best For | Starts With | Output |
| -------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------- | -------------------------------------- | ------------------- |
| [**Cinematic Lab**](/features/studio/cinematic-lab) | Creating cinematic master stills, lookdev frames, key art, concept images, and source frames for video generation | Text prompt, optional reference images | Image grid |
| [**Multi-Cam**](/features/studio/multi-cam) | Generating alternate camera angles from a still or clip, building coverage, creating cinematic grids | Image or video | Image grid or video |
| [**Motion Director**](/features/studio/motion-director) | Animating a still with a controlled camera move: push, pull, orbit, dolly, handheld, crane, drone, and more | Image | Video |
| [**AI Transitions**](/features/studio/ai-transitions) | Bridging a start frame and end frame into a seamless moving transition | Two images | Video |
| [**Avatar Studio**](broken://pages/wTvU4zCSlGxhDtkeOPbJ) | Generating talking-head presenter videos from a photo or script. Best for UGC ads, spokesperson clips, and social content. | Photo + script or prompt | Video |
**How to think about Production workflows**
Use **Cinematic Lab** when you need the hero frame. Use **Multi-Cam** when you need coverage around that frame. Use **Motion Director** when one still deserves motion. Use **AI Transitions** when two images need to become one shot.
A strong Studio production chain often looks like this:
1. Create a hero still in **Cinematic Lab**
2. Generate alternate angles with **Multi-Cam**
3. Animate the strongest frame in **Motion Director**
4. Bridge two moments together with **AI Transitions**
***
#### Post-Production Department
Post-Production workflows refine, repair, transform, or finish footage you already have.
| Workflow | Best For | Starts With | Output |
| ----------------------------------------------------------- | ----------------------------------------------------------------------------------------------- | ------------------------------ | ---------------------------- |
| [**Rotoscope**](/features/studio/sam-3-rotoscoping) | Isolating subjects and separating foreground from background | Video | Edited video / cutout result |
| [**Erase Objects**](/features/studio/object-eraser-tool) | Removing unwanted objects, people, signs, gear, or distractions from footage | Video | Cleaned video |
| [**Add Effects**](/features/studio/kling-vfx) | Adding VFX such as fire, rain, fog, energy, atmosphere, or stylized scene changes | Video | VFX video |
| [**Reshoot**](/features/studio/reshoot) | Making targeted retake-style changes to parts of a clip | Video | Edited video |
| [**Upscale**](/features/studio/video-upscaling) | Increasing resolution and improving the finish of a clip | Video | Higher-resolution video |
| [**Motion Capture**](/features/studio/kling-motion-control) | Transferring motion from a reference video onto a character image | Motion video + character image | Video |
| [**Relight Scene**](/features/studio/relight-scene) | Changing light direction, mood, or atmosphere on an image or short clip | Image or video | Relit image or video |
| [**Reframe**](broken://pages/ttgYJltho5RdTpEtH0fq) | Changing a clip's aspect ratio with AI edge fill. Best for repurposing 16:9 as 9:16 for social. | Video | Reframed video |
**How to think about Post-Production workflows**
Use **Rotoscope** when you need separation. Use **Erase Objects** when something should disappear. Use **Reshoot** when a specific part of the shot should change. Use **Add Effects** when the scene needs new visual energy. Use **Upscale** at the end, not the beginning, after the creative changes are already approved.
***
#### Audio Department
The Audio department is visible in Studio so you can see where sound workflows will live, but the current Audio cards are not available yet.
Coming Soon.
{% hint style="warning" %}
**Coming Soon cards are previews, not active workflows.** If a card is locked, it is intentionally unavailable in this release.
{% endhint %}
***
#### Choosing the Right Workflow
Goal
Start Here
I need a cinematic still from an idea
Cinematic Lab
I need more angles from one image or clip
Multi-Cam
I want to animate a still
Motion Director
I need a transition between two frames
AI Transitions
I need to isolate a subject
Rotoscope
I need to remove something from footage
Erase Objects
I want to add rain, fire, fog, energy, or stylized VFX
Add Effects
I need a targeted retake or scene change
Reshoot
I want to improve resolution after the edit is approved
Upscale
I need to transfer motion onto a character
Motion Capture
I need to change the light or mood of a scene
Relight Scene
I want to change a clip's aspect ratio for social
Reframe
I need a talking-head presenter video from a photo or script
Avatar Studio
***
#### Pro Workflows to Try First
**Concept frame to moving shot**
Use this when you need a shot that never existed.
1. Open **Cinematic Lab**
2. Generate a 4-up batch of possible hero frames
3. Pick the strongest still and click **Done**
4. Open **Motion Director**
5. Use that still as the source image and choose a camera movement
This is the cleanest way to go from idea to usable video: design the frame first, then animate it.
***
**Missing coverage from an existing edit**
Use this when your timeline has the moment, but not the angle.
1. Park the Premiere playhead on a useful frame
2. Capture that frame into **Cinematic Lab** or **Multi-Cam**
3. Ask for a new angle, insert shot, detail, or variation
4. Reuse the result from Recents in another Studio workflow if needed
This is especially helpful for interviews, product videos, real estate, documentary edits, and any project where you need one more shot after production is over.
***
**Clean, transform, finish**
Use this when you have real footage but it needs help.
1. Use **Erase Objects**, **Reshoot**, or **Add Effects** for the creative change
2. Review the result in chat or the video editor surface
3. Only after the creative version is approved, run **Upscale**
Do not upscale first unless resolution is the only job. Upscaling an intermediate clip wastes time and can make later AI passes less flexible.
***
#### Best Practices
**Start with the workflow, not the model**
Studio exists so you do not have to memorize model names. Pick the creative job first. The workflow will expose the controls that matter for that job.
**Use Recents to chain workflows**
The fastest Studio users do not constantly upload and download files. They generate, click **Done**, then reuse the result from **Recents** in the next workflow.
**Keep references organized**
For visual consistency, build a small set of repeatable references: hero frames, product shots, character portraits, location stills, and lighting examples. Reuse them across Cinematic Lab, Multi-Cam, Motion Director, and Relight Scene.
**Generate stills before video when the look matters**
Video generation is more expensive and less forgiving than image generation. If the framing, subject, lighting, or wardrobe matters, lock the still first in Cinematic Lab, then animate it.
**Use Upscale last**
Upscale is a finishing pass. Run it after the clip is creatively approved, not before every experiment.
**Pay attention to asset requirements**
If a workflow asks for a still, give it a clear still. If it asks for video, give it the shortest clip that contains the motion or shot you need. Cleaner inputs make every AI pass more predictable.
***
#### Troubleshooting
**I do not see a workflow I expected**\
Use the Studio search box and try the job name instead of the model name. For example, search `remove`, `rotoscope`, `angle`, `transition`, `relight`, or `upscale`.
**A card is visible but locked**\
That workflow is Coming Soon. Locked cards are previews of planned Studio departments, not active tools.
**The asset loader is not showing the file type I want**\
The workflow may not support that asset type. Motion Director only accepts images. Rotoscope only accepts video. Motion Capture needs one video and one image. The loader filters inputs to prevent unsupported jobs from starting.
**My workflow opened the video editor instead of staying in Studio**\
Some post-production tools use the full video editor surface because they need masking, preview, or timeline-style controls. That is expected for workflows like Rotoscope, Erase Objects, Add Effects, Reshoot, and Upscale.
**Generation fails before it starts**\
Check Settings and confirm your FAL API key is configured. Many Studio workflows use cloud models through the local Chat Video Pro service.
**I lost where I was**\
Press **Escape** once to return to the Launchpad from a workflow. Press **Escape** again to close Studio and return to chat.
***
**Next:** Start with Cinematic Lab to create a cinematic still, then use that still in Motion Director, Multi-Cam, AI Transitions, or Relight Scene.
# Cinematic Lab
Cinematic Lab is a still-image workflow that lets you describe a scene, choose the camera body, lens, focal length, and aperture you'd shoot it on.
{% hint style="info" %}
**Cinematic Lab makes still images, not video.** This is the photography stage of your pipeline. Once you have a still you love, take it into **Motion Director** to animate it, **Multi-Cam** to generate alternate angles, or **AI Transitions** to bridge two stills into a moving shot.
{% endhint %}
#### When to Use Cinematic Lab
Cinematic Lab is the right tool any time you'd rather **describe** a frame than shoot it — and you want it to look like a frame from a movie, not a stock render.
* **Concept frames and mood boards** — pitch a look before you book talent, locations, or gear
* **AI key art and thumbnails** — generate cinematic stills sized to whatever aspect ratio your edit needs
* **Pre-vis stills for an edit you haven't shot yet** — drop a placeholder that already matches the lens and lighting plan
* **Source frames for video generation** — Cinematic Lab → Motion Director, or Cinematic Lab → AI Transitions, lets you start any motion workflow from a still you actually like
* **Reference-driven scene matching** — feed in a still from your edit and ask Cinematic Lab to render the next angle, alternate beat, or styled variation
* **Title cards, lookdev plates, story frames** — any time the brief is "make this feel like a real frame from a real production"
{% hint style="warning" %}
**Cinematic Lab is not a chatbot or a Premiere automation.** It runs as a standalone Studio workflow. When you click Done, your selected images are inserted into the chat as a normal image message — from there you can drag to your project, save them to Library, or feed them into another Studio workflow. Cinematic Lab does not edit footage on your timeline.
{% endhint %}
#### Getting Started
**Step 1: Open Cinematic Lab**
1. Open Chat Video Pro (Window → Extensions → Chat Video Pro)
2. Click the **Studio** button in the sidebar to open the Launchpad
3. In the **Production** department, click the **Cinematic Lab** card
Cinematic Lab is one of the few Studio workflows that **does not** open the asset loader first — because it generates from a written description rather than transforming an existing clip. You'll go straight into Step 1: Describe Your Scene.
***
**Step 2: Describe Your Scene**
This is the most important box on the page. The model treats your description as the **subject and action** layer of the prompt — everything visual (camera, lens, lighting, photorealism anchors) gets layered on top in Step 3, so you don't need to write technical language here. You need to describe **what's happening, who's in frame, and where they are**.
**Three things to always include:**
What
Example
Subject
"A woman in a wool coat", "An empty diner booth", "A weathered fisherman"
Action / state
"lighting a cigarette", "mid-laugh", "staring out the window"
Setting and time
"in a fog-soaked harbor at dawn", "in a neon-lit ramen shop after midnight"
**Optional but powerful:**
* Mood and emotional tone ("melancholy", "tense", "intimate")
* Wardrobe, props, and color palette ("amber Carhartt jacket, oxidized brass rings")
* Weather and atmosphere ("rain, blown sideways", "thick fog catching the headlights")
* Composition direction ("close-up on hands", "wide shot, subject lower-third")
**Example prompt:**
> "A detective standing in a rain-soaked alley at night. Trench coat dripping, neon signs reflecting in the puddles, shoulder of a passing pedestrian blurred in the foreground. He's mid-thought, looking at something off-camera."
{% hint style="info" %}
**Fastest way to prompt:** Use the microphone button to dictate the scene out loud. Cinematic prompts come out better when you describe a moment instead of a list — speaking is naturally narrative.
{% endhint %}
#### AI Optimize ✨
Tap the sparkle (✨) button next to the scene description to get an AI-improved version of your prompt. The optimizer understands your scene description and any reference images you have attached — avoiding camera and lens language that is already set in a later step, so it focuses on subject, action, setting, and mood. Choose **Replace** to apply it, **Regenerate** to try again, or **Close** to keep your original.
**Reference Images (Up to 14)**
Can attach up to **14 reference images** from any of three sources:
* **Upload** — JPEG, PNG, or WebP from your computer
* **Frame capture** — pulls the current frame from your active Premiere Pro timeline (great for matching a look that already exists in your edit)
* **Recents** — pick from any image generated, captured, or used recently across Chat Video Pro. Multi-select is enabled so you can grab a whole mood board at once.
References change what Cinematic Lab is doing under the hood. With **no references**, the model generates from scratch using your description plus the camera/lens settings. With **references attached**, the model treats them as a style and identity guide — matching wardrobe, faces, lighting palette, art direction, and texture — while still applying your selected camera and lens characteristics.
**What to use references for:**
Goal
What to attach
Match an actor's face across shots
One or two clean head-and-shoulders portraits of the subject
Match a wardrobe or prop
A close-up of the costume / item from any angle
Match a location's color and lighting
One or two stills from the location at the right time of day
Match a film's overall look
3–6 frames from a reference film, all from the same chapter or mood
Generate the next angle from your edit
Capture a frame from your timeline and ask for "the same character from a low angle"
**Step 3: Choose Your Look**
This is the page that separates Cinematic Lab from a normal text-to-image generator. Instead of dumping technical jargon into your prompt, you make four choices the way a DP would. The system translates each one into specific visual language the model has been trained to interpret physically — not as filters, but as optical consequences.
**The four columns**
* **Camera Body** — the *sensor and color science*. Different cameras render skin tones, highlights, contrast, and grain in fundamentally different ways.
* **Lens** — the *character of the glass*. Modern primes are clean and clinical; vintage lenses bloom and flare; anamorphics give you oval bokeh and 2.39:1 widescreen feel.
* **Focal Length** — *how the world is compressed*. 24mm exaggerates space; 50mm is the human eye; 135mm flattens everything for portrait compression.
* **Aperture** — *how much is in focus*. f/1.4 is paper-thin; f/8 is everything sharp from foreground to horizon.
You're not picking metadata — you're picking the lens you would have shot on. Underneath, the prompt engine assembles a four-pillar prompt that puts technical setup first (so the model treats your gear choice as the *visual container* before it draws anything) and adds photorealism anchors at the end (so it doesn't drift into illustration).
***
**Camera Body — pick the look, not the brand**
Night scenes, urban exteriors, modern commercial, documentary. Best when you want everything legible in the dark.
RED V-Raptor
Clinical sharpness, 8K-feel detail, punchy reds, high contrast, "digital" precision
Sci-fi, action, VFX-heavy, commercials with crisp products. The opposite of soft.
Blackmagic Pocket 6K
High saturation, thick color density, gritty textured grain, raw indie energy
Music videos, indie shorts, anything that should feel a little unpolished and alive.
Canon C500 Mark II
Warm magenta-pink skin bias, soft warm highlights, organic but sharp
Portraits, interviews, intimate drama, any human-centric frame where the face is the subject.
***
**Lens — pick the personality**
Lenses are split into three categories. The category alone changes the **shape of light** in the frame.
**Spherical (modern, clean)**
Lens
Best For
Zeiss Supreme Prime
Clinical perfection. Modern features, sci-fi, anything where "razor-sharp" is a virtue.
The shallow-focus default — most narrative work lives here
f/2.8
Professional balance, slight environmental context, sharp on subject
Commercial work, two-shots where both subjects need to be sharp
f/4.0
Moderate depth, sharper overall, more world visible
Storytelling shots where the location matters as much as the subject
f/8
Deep focus, sharp foreground to background, hyperfocal
Landscapes, architectural frames, anything where the whole scene is the subject
***
**Output Controls — Quantity, Aspect Ratio, Resolution / Quality, and Model**
Below the four selectors is your output bar.
* **Quantity (1–4)** — generate up to 4 variations in one batch. With Nano Banana 2 each generation is fast enough that you should default to 4 for first passes — pick a winner, regenerate with adjustments, narrow down.
* **Aspect Ratio** — 10 options including 16:9 (landscape default), 9:16 (vertical), 1:1, 21:9 (cinematic), 4:3, 3:2, 2:3, 5:4, 4:5, 3:4. Pick the ratio of the **deliverable**, not the camera — the model renders a real frame for that shape.
* **Resolution / Quality** — what shows here depends on the model you choose:
* **Nano Banana 2** exposes **1K, 2K, 4K** as a resolution dropdown. Default 2K; bump to 4K when you're picking finals or generating key art.
* **GPT Image 2** exposes a **quality** dropdown (Low / Medium / High) instead. High = slowest, sharpest, most detailed.
* **Model picker** — a small dropdown lets you switch between the available image models. (See **Choosing the Right Model** below.)
When everything's set, click **Generate**. You'll move to Step 4: Results.
{% hint style="warning" %}
**Regenerating wipes existing results.** If you've already generated a batch and you change anything in Step 3, hitting **Generate** again replaces the previous set. If there's a frame you want to keep, hit **Done** on it first (which inserts it to chat) before you regenerate.
{% endhint %}
#### Choosing the Right Model
**Nano Banana Models**
Google's latest image model.
* **Fast** is the default. Twice as fast as the previous generation, half the price, supports up to 4K resolution, and is good enough for \~90% of work.
* **Pro** adds extended reasoning and web search. Use it for harder prompts: rare locations, real-world references, complex multi-subject scenes, or any prompt where the model needs to "think" about what something actually looks like.
**Best at:**
* Photorealism — texture, grain, skin pores, lighting physics
* Reference-guided generation (reference images influence both content and look)
* High-resolution output (4K natively)
* Speed — first-pass exploration is dramatically faster than alternatives
**Choose Nano Banana 2 when:** you're doing most of your work. It's the right default for portraits, landscapes, character work, mood pieces, anything reference-driven, and any frame headed for video generation in another Studio workflow.
**GPT Image 2**
OpenAI's high-fidelity image model.
**Best at:**
* **Text in the image** — readable signs, posters, book covers, neon, packaging, type-driven shots. This is the single biggest reason to switch off Nano Banana.
* **Complex multi-subject scenes** — three people in a room, a busy street, a crowded restaurant
* **Prompt adherence on weird, specific requests** — surreal compositions, hard logical setups, unusual perspectives
* **Conceptual / illustration-leaning frames** — when "real photograph" isn't the only goal
**Choose GPT Image 2 when:** there's text in the frame, the scene has a lot of moving parts, or the prompt is conceptually weird. It's slower and more expensive than Nano Banana 2 Fast — but it earns it on the prompts it's better at.
{% hint style="info" %}
**Pro tip — generate the same prompt on both models.** When you're locking in the look for a project, run the same scene through Nano Banana 2 and GPT Image 2 once. Side-by-side, you'll know within thirty seconds which model owns *this* project's aesthetic. Then commit and stop second-guessing.
{% endhint %}
***
#### Working With Your Results
* **Click an image** to open it fullscreen for inspection. Click the backdrop or close button to dismiss.
* **Click the checkbox in the corner** to add or remove an image from your final selection. The first image is pre-selected by default.
* **Regenerate** runs the same configuration again to produce a new batch. (See the warning above about wiping previous results — if you want to keep a frame, click **Done** on it before regenerating.)
* **Done** inserts your selected images into the chat as a normal image message and closes Cinematic Lab.
**What "Done" Actually Does**
Selected images are added to the chat as an image grid (the same layout used everywhere else in Chat Video Pro for multi-image generations) and saved to the Library with full metadata — your scene description, camera, lens, focal length, aperture, model, and resolution all stay attached. From the chat message you can:
* **Download the image to use** — Imports directly into Premiere Pro bins
* **Pull it back into another Studio workflow from Recents** — it'll show up in the Recents picker for Motion Director, Multi-Cam, AI Transitions, Relight Scene, and the next Cinematic Lab session
***
#### Pro Tips
**Pair the camera with the lens like you're packing a kit, not picking from a list**
Don't shop the dropdowns alphabetically. Decide what the project should *feel* like, then pick a body and lens that already work together in the real world. ARRI + Cooke is prestige drama. Sony + Zeiss is modern commercial. RED + Sigma is "shot yesterday." Blackmagic + K35 is grungy and nostalgic. Canon + Canon is the warmest human-centric pairing in the lab. The model has been trained on actual footage from these combinations — leaning into a real-world pairing produces noticeably more cohesive frames than mixing randomly.
**Lead with subject, not gear**
The camera/lens controls are doing the technical work for you. Don't waste your scene-description box repeating "shot on Alexa, 50mm, f/2.0" — you've already selected that. Use those words for *who* and *what*. The four-pillar prompt engine layers the technical pillar first under the hood, and the model treats it as the visual container before it draws your subject.
**Use 4-up batches as a focusing exercise**
Default to quantity 4 for a first pass on any new prompt. Don't pick the best one — pick the *direction* the best one is pointing. Then change one thing (a different lens, a different aperture, a slightly tweaked description), generate 4 more, and compare. You'll converge on a final frame in 2–3 batches that would have taken 15 single-shot generations.
**Reference images are the fastest way to character-consistency**
If your project has a recurring character, location, or product, attach 1–2 clean reference images or an element on every generation in that project. Nano Banana 2 reads references as both *what to render* and *how it should look* — you'll get cross-shot consistency that's nearly impossible from text alone. For projects with a returning subject, drop a portrait reference into a Library collection once and re-attach it from Recents every session.
**Match-the-edit workflow: Frame Capture into Cinematic Lab**
If you've already shot something and need a missing angle, an alt take, or a stylized variation, click **Frame Capture** in Step 1 to pull the current frame from your Premiere timeline as a reference. Now describe the variation you want ("same character, low angle from below") and the model will use the captured frame as the identity and lighting anchor while honoring your new prompt and gear choices. This is the single fastest path from a real edit to a believable AI-generated companion frame.
**Bridge Cinematic Lab into a moving shot**
A still you love isn't the end — it's the start. Once you have a hero frame, hit **Done** to drop it into chat, then:
* Drag it into **Motion Director** to animate it with a camera move (push in, orbit, dolly back)
* Drag it into **Multi-Cam** to generate alternate angles from the same instant
* Pair it with another still in **AI Transitions** to interpolate motion between them
* Drag it into **Relight Scene** to change the lighting mood while keeping composition
Cinematic Lab is the photography stage — it's most powerful when you treat it as the entry point to the rest of the Studio.
**Try GPT Image 2 the moment you need text in the frame**
Nano Banana 2 will hallucinate signs, packaging, and titles into something that *looks* like text but isn't. The instant your scene includes a readable word — a license plate, a billboard, a book cover, a chyron — switch the model picker to GPT Image 2. It's the only model in the picker that handles real text reliably.
**Pro mode is for hard prompts, not all prompts**
Nano Banana 2 Fast is the right default. Save Pro for prompts that need *thinking* — rare real-world locations, weird logical setups, scenes where the relationship between elements has to be physically consistent. For everyday hero frames, Fast is genuinely better because the speed lets you iterate more.
**Build a project lookbook once**
The most consistent-looking AI projects come from the same trick every time: pick your camera + lens + 4–6 reference images at the start of the project, save them somewhere (a Library collection, a folder, even a chat message), and reuse the exact same set on every Cinematic Lab generation for that project. You're not generating images — you're operating a virtual production with consistent gear and art direction.
***
#### Troubleshooting
**Generations look painted, plastic, or "AI-y"**\
This is illustration drift. Try to drop the resolution to 2K (4K can over-smooth at certain prompts), and add concrete physical details to your scene description — *"stubble, sweat on his temple, dust on the lapel"* gives the model the textures it needs to stay grounded.
**Faces look generic or keep changing**\
Add a reference image. Even one clean head-and-shoulders portrait will lock the face. For projects with a recurring character, drop the reference into Recents once and re-attach it on every generation.
**The lens choice doesn't seem to be doing anything**\
Two likely causes. First — your scene description is fighting it. If you ask for "everything in focus" while choosing f/1.4, the model has to compromise. Match your description to your selected aperture, or remove focus language and let the gear do the work. Second — your prompt is too short. The lens characteristics layer in *underneath* a real subject; "a guy" plus a Cooke S4/i won't show off the Cooke Look. Give the model something to render before you expect lens character to appear.
**Text in the image is gibberish**\
Switch the model picker to GPT Image 2. Nano Banana 2 is fast and beautiful but it will not reliably render real words.
**Generations are too saturated, too HDR, too "Marvel"**\
Try a different camera body. Avoid RED if you don't want punchy contrast. ARRI and Canon C500 are your most natural-looking bodies. You can also bump the aperture from f/1.4 to f/2.0 or f/2.8 — extreme shallow focus can read as artificial.
**My references are influencing too much (everything looks like the reference, ignoring my prompt)**\
You probably have too many references attached, or they're too visually similar. Drop to 2–3 references and make sure your scene description is detailed enough to give the model something *new* to construct.
**My references are influencing too little (the output ignores them)**\
Nano Banana 2 handles references better than GPT Image 2 — switch models. Also check that your description doesn't directly conflict with the reference (asking for "blonde hair" when your reference has black hair forces the model to choose).
**I keep losing my favorite frame when I regenerate**\
Always click **Done** on a frame you want to keep *before* hitting **Regenerate**. Done inserts it into chat where it's safe; Regenerate replaces the current results grid. But all generations are stored in your Fal library even if not saved locally
**The Generate button is disabled or generation fails immediately**\
Cinematic Lab needs your FAL API key configured in Settings to generate. If it's missing or invalid, generation will fail before it starts. Set it under Settings → API Keys. Also check that it is funded.
***
**Next:** Learn about Motion Director — animate any Cinematic Lab still into a moving shot with a chosen camera move.
# Multi-Cam
Multi-Cam creates new camera angles from an existing image or video. Use it when you have the subject, scene, or performance you want, but need more coverage: a side profile, 3/4 angle, low angle, high angle, over-the-shoulder shot, wide shot, character sheet, or new video angle that keeps the style of the original clip.
{% hint style="info" %}
**Multi-Cam is a coverage tool.** It works best when the source already contains the subject, styling, wardrobe, lighting, and world you want to preserve. You are asking for a new camera position, not a new scene.
{% endhint %}
#### When to Use Multi-Cam
Use Multi-Cam when you need additional angles from source material you already like.
* **Create alternate still angles** from a hero image, product frame, portrait, generated character, or concept shot
* **Build a character sheet** from one strong character image
* **Generate 3x3 cinematic grids** to explore shot language quickly
* **Create profile, 3/4, high, low, and overhead views** for visual planning
* **Generate reverse or side coverage** from a short video clip
* **Turn one-camera footage into edit options** for interviews, product demos, narrative scenes, and b-roll
* **Prepare references for other Studio workflows** like Motion Director, AI Transitions, Relight Scene, and Cinematic Lab
Multi-Cam is especially useful when you need options before the edit is locked. Instead of asking one model for "the perfect shot," you can generate a controlled spread of angles and choose the frames that actually cut.
***
#### Image vs. Video
Multi-Cam has two different paths depending on the asset you start with.
Input
What You Can Generate
Best For
Image
Up to 4 single camera angles, or one 3x3 grid preset
Short clips, interviews, performance coverage, b-roll variations
The image path is broader and more exploratory. The video path is more focused because it has to preserve motion, timing, and audio continuity.
***
#### Getting Started
**Step 1: Add a Source Asset**
Multi-Cam starts with the Studio asset loader and accepts image or video.
You can start from:
* **Upload**
* **Recents**
* **Frame capture / timeline sources supported by Studio**
For images, the first image becomes the primary source. You can also attach more image references in the composer after the workflow opens.
For video, the video is the source clip and the composer locks the media strip to that one clip. This prevents accidentally mixing multiple videos or adding image references to a path that expects one video.
{% hint style="warning" %}
**Start with a clean source.** Multi-Cam can create new angles, but it still depends on the original subject being readable. Blurry faces, motion blur, heavy compression, cropped bodies, and unclear products all make angle generation harder.
{% endhint %}
### Image Workflow
The image workflow is designed for controlled still-angle generation.
1. Add a source image
2. Choose up to **4 camera angles**, or choose **1 grid preset**
3. Add optional notes in the composer
4. Set output aspect ratio and quality
5. Generate
6. Select the results you want to save
7. Optionally upscale selected images
8. Click **Done** to add them to chat
#### Image Preset Types
Multi-Cam includes **13 image presets**:
Grid presets are mutually exclusive with single angles. If you select a grid, you generate the grid. If you select single angles, you can select up to four.
***
#### Grid Presets
Grid presets are for fast exploration. They generate one full grid, then Multi-Cam splits it into individual selectable cells.
Grid
Best For
What It Produces
Cinematic Grid
Shot exploration, camera planning, visual options
A 3x3 contact sheet of cinematic angles
Character Sheet
Character design, identity references, consistency planning
A 3x3 reference sheet with front, side, back, close-up, and hero views
**Cinematic Grid**
Use Cinematic Grid when you want to see several camera possibilities at once.
It is useful for:
* Planning coverage
* Finding a hero angle
* Creating references for a pitch deck
* Testing whether a character or product works from multiple viewpoints
* Building a set of stills to feed into AI Transitions or Motion Director
**Character Sheet**
Use the Character Sheet when the subject needs to stay consistent across future generations.
It is useful for:
* Character development
* Reference sheets or elements
* Costume and identity checks
* Multi-angle prompts in later workflows
{% hint style="info" %}
**Pro tip: save both the grid and the best cells.** In the results screen, you can select individual cells and also use **Save Cinematic Grid** to keep the full grid image. The grid is useful as a reference board, while individual cells are easier to reuse in other workflows.
{% endhint %}
#### Geometric Angles
Geometric angles are for clear camera repositioning. They are best when you want the model to change perspective while preserving the subject.
Angle
Best For
Notes
Head-On
Direct portraits, product front views, clean reference images
Use when you need a neutral front-facing image
Profile Left
Side references, character turnarounds, interview coverage planning
Use geometric angles when the goal is practical coverage. For example, if a product is only shown from the front and you need a side view, choose Profile Left or Profile Right instead of a stylized preset.
***
#### Stylistic Angles
Stylistic angles are more cinematic and interpretive.
Angle
Best For
Notes
Over Shoulder
Dialogue, interviews, character interaction, narrative framing
Adds a blurred foreground shoulder and focuses on the subject
Dutch Angle
Tension, unease, music videos, thriller tone
Tilts the camera while trying to preserve anatomy
Wide Shot
Establishing frames, environment reveal, scale
Pulls back to show more of the world
Use stylistic angles when the shot needs a stronger editorial feeling, not just a new view.
{% hint style="warning" %}
**Stylistic angles invent more.** Over Shoulder, Dutch Angle, and Wide Shot can add composition and environment that were not fully visible in the source. Use them when that extra interpretation is useful, not when you need exact technical continuity.
{% endhint %}
#### Composer Controls for Images
The composer lets you add steering notes and reference images.
**Notes**
Use the notes box for shot direction, not a full rewrite.
Good notes:
* "Keep the same wardrobe and background."
* "heroic pose, dramatic lighting."
* "clean commercial product photography"
* "Make the subject full body."
* "Preserve the sci-fi cockpit environment."
* "Keep the original expression and hair."
Weak notes:
* "Make it better."
* "more cinematic"
* "change everything"
* "new outfit, new pose, new location."
**Reference Images**
Multi-Cam image workflows support a multi-image strip. You can add images from upload, recents, frame capture, or saved Elements through the composer.
Use extra references when:
* The source image does not show the full outfit
* A face or character needs stronger consistency
* A product needs extra material or branding detail
* You want the model to understand the subject from more than one view
The first image is the primary source. Additional images are references, not a replacement for the source.
**Aspect Ratio**
Image outputs can use:
Option
Use For
Portrait (9:16)
Shorts, Reels, TikTok, vertical thumbnails
Portrait (3:4)
Character sheets, portraits, reference images
Landscape (16:9)
YouTube, cinematic frames, widescreen b-roll
Multi-Cam detects the source shape and sets an automatic starting point, but you can override it.
**Quality**
Use **Standard** while exploring angles. Use **High Quality** when you know which angle you want and need a stronger final image.
High Quality takes longer, but can improve detail and final polish.
***
### Video Workflow
The video workflow creates a new moving angle from a source clip.
1. Add a source video
2. Choose one supported video angle
3. Add optional notes
4. Choose duration: **5s**, **10s**, or **15s**
5. Choose whether to keep the audio
6. Generate
7. Preview the new angle
8. Click **Done - Add to Chat**
Video Multi-Cam uses **Kling O3 Pro Multi-Cam** through the O3 video-to-video reference endpoint.
#### Supported Video Angles
The video path is intentionally limited to the most reliable camera moves.
Angle
Best For
Profile Left
Side coverage from camera-left
Profile Right
Side coverage from camera-right
3/4 Left
Natural alternate angle with subject depth
3/4 Right
Natural alternate angle in the opposite direction
Low Angle
Heroic or more powerful shot
High Angle
Elevated or more observational shot
Grid presets, Bird's Eye, Over Shoulder, Dutch Angle, Wide Shot, and Head-On are not shown in the video path because they require more invention than the current video workflow is designed to preserve.
{% hint style="info" %}
**Pro tip: Video Multi-Cam is best for clean camera repositioning.** If you need a surreal angle, a completely new environment, or a heavy stylized shot, generate a still angle first, then use Motion Director to animate it.
{% endhint %}
#### Duration
The video path supports **5**, **10**, and **15** second outputs.
Duration
Best For
5s
Quick coverage, reaction inserts, tests
10s
Interview coverage, b-roll, product moments
15s
Longer action where the subject remains stable
Longer outputs take more time and give the model more motion to preserve. Use 5 seconds for testing and 10-15 seconds for approved ideas.
#### Audio
The video path includes an audio toggle:
Option
Use When
Keep
You want the original audio carried into the generated angle
Remove
You are cutting to music, doing sound design separately, or only need visuals
For most editorial work, use **Keep** when generating interview or dialogue coverage and **Remove** when generating b-roll or visual-only cutaways.
***
#### Working With Results
**Image Results**
After image generation, Multi-Cam shows an image grid.
You can:
* Select individual results
* Select all results
* Upscale selected images
* Save the full cinematic grid when using a grid preset
* Regenerate
* Click **Done** to add selected images to chat
If one angle fails but others succeed, Multi-Cam keeps the successful results and warns you that some angles failed. This is useful when generating several angles at once, because one bad angle does not discard the whole batch.
**Video Results**
After video generation, Multi-Cam shows a video preview.
You can:
* Play the result
* Generate another angle
* Add the video to chat
The saved video message stores the Multi-Cam source, angle ID, angle name, duration, audio setting, and aspect-ratio metadata for the chat player.
***
#### Pro Tips
**Use 3/4 angles first**
3/4 Left and 3/4 Right are often the safest alternate angles. They change perspective without forcing the model into a full side profile or extreme camera position.
**Use profile angles when you truly need a side view**
Profile Left and Profile Right are practical, but demanding. They work best when the subject is clearly visible and the source image has enough detail for the model to understand face, clothing, and silhouette.
**Use grids for exploration, not final selection**
Cinematic Grid and Character Sheet are great for finding options. After you see which cell works, save that cell and use it as the source for a more focused generation.
**Attach references when identity matters**
If a character, product, mascot, or wardrobe detail must stay consistent, add extra references in the composer. One source image may not contain enough information for every new angle.
**Do not ask for a new story in the notes**
Notes should support the camera angle. "Keep the same jacket and neon street" is useful. "Put them in a castle with a new outfit" fights the purpose of Multi-Cam.
**Use High Quality after you choose the angle**
Generate at Standard to compare angles quickly. Once you know the angle, regenerate or upscale selected images for final use.
**For video, choose a stable source clip**
Video Multi-Cam works best when the clip has one clear subject, consistent lighting, and simple movement. Fast cuts, heavy motion blur, or multiple competing subjects make the new angle less reliable.
**Think like an editor**
Ask what shot is missing from the cut: a reaction, profile, high angle, low angle, detail shot, wide, or reference sheet. Multi-Cam works best when the target angle has a job.
Example Workflows
**Character Turnaround From One Hero Image**
Setting
Choice
Input
Character portrait or full-body frame
Presets
Profile Left, Profile Right, 3/4 Left, 3/4 Right
Aspect Ratio
Portrait (3:4)
Quality
Standard first, High Quality for keepers
Notes
"preserve exact outfit, hair, and facial structure"
**Cinematic Grid for Shot Planning**
Setting
Choice
Input
Hero frame or concept image
Preset
Cinematic Grid
Notes
"professional sci-fi action coverage, maintain the same environment"
Result
Save the full grid and the strongest individual cells
**Product Angle Set**
Setting
Choice
Input
Product front view
Presets
3/4 Left, 3/4 Right, Profile Left, Wide Shot
Aspect Ratio
Landscape (16:9) or Portrait (3:4)
Notes
"clean commercial product photography, preserve logo and material finish"
**Interview Reverse Coverage**
Setting
Choice
Input
Short interview clip
Preset
3/4 Left, 3/4 Right, Profile Left, or Profile Right
Duration
5s or 10s
Audio
Keep
Notes
"maintain the original lighting and background, continue the speaking motion"
**B-Roll Variation From One Clip**
Setting
Choice
Input
Product or lifestyle b-roll clip
Preset
Low Angle or High Angle
Duration
5s
Audio
Remove
Notes
"keep the same product and motion, make the angle feel like complementary coverage"
***
#### Best Practices
**Separate image planning from video generation**
If you are unsure which camera angle you want, test angles on a still first. Once the visual direction is clear, generate video coverage.
**Preserve screen direction**
Left and right matter in edits. If a character is looking camera-right in the source, think about which alternate angle will preserve the intended eyeline or reverse it.
**Avoid tiny subjects**
If the subject is too small in frame, the model has less identity information. Use Wide Shot only when the source has enough detail or when exact face fidelity is less important.
**Avoid text-heavy sources**
Fine text, logos, UI, and product labels may drift. If branding must remain exact, keep expectations realistic and inspect results carefully. You can also attach up to 14 reference images for extra detail for the model.
**Use generated angles as options, not truth**
Multi-Cam produces AI-generated coverage. Treat it like a creative angle pass, then choose the frames that serve the edit.
***
#### Troubleshooting
**The Generate button is disabled**\
Select at least one angle or one grid preset. For video, select one video angle. Also confirm your FAL API key is configured.
**Only some image angles finished**\
Multi-Cam keeps successful angles even if one fails. Save the successful images, then regenerate the missing angle separately.
**The face changed too much**\
Attach more reference images or use less extreme angles. 3/4 angles are usually safer than full profiles, Bird's Eye, or Wide Shot.
**The profile angle still looks too frontal**\
Use Profile Left or Profile Right with notes like "pure side profile, no eye contact, no front-facing pose." Source images with clear face structure work better.
**The grid has weak cells**\
That is normal for exploratory grids. Save the best cells and regenerate focused single angles from those cells.
**The video angle does not match the original motion**\
Try a shorter source clip with simpler action. Video Multi-Cam has to preserve motion and create a new camera position at the same time.
**The video generation takes several minutes**\
Kling O3 Pro Multi-Cam can take a few minutes, especially for longer durations. Keep the panel open until the result appears.
**The output framing is not what I expected**\
For image results, set the output aspect ratio before generating. For video results, the model follows the source video and selected angle more than the image aspect controls.
***
**Next:** Use Motion Director to animate a generated still angle, use AI Transitions to bridge two Multi-Cam frames, or use Relight Scene to match lighting before generating coverage.
# Motion Director
Motion Director animates a still image with a directorial camera move. Instead of writing a vague prompt like “make this cinematic,” you choose the shot language: dolly in, dolly out, orbit, vertigo, tracking shot, crane up, aerial reveal, FPV drone, or handheld. Motion Director analyzes the source image, builds a movement-specific Kling 3.0 Pro prompt, and returns a finished video you can add back to chat and use in Premiere.
{% hint style="info" %}
**Motion Director is for camera movement, not full scene replacement.** The goal is to preserve the image you already have while adding motion. If you need to create the still first, start in Cinematic Lab. If you need alternate angles before animating, use Multi-Cam. If you need to change the lighting, use Relight Scene.
{% endhint %}
#### When to Use Motion Director
Use Motion Director when you have a still frame that already works — composition, character, product, or location — and you need to animate it.
* **Animate a Cinematic Lab frame** into a real video beat
* **Turn a thumbnail or key art still** into a teaser, intro, ad, or title-card motion shot
* **Add subtle camera movement** to static product shots, real estate stills, concept art, and hero frames
* **Create missing motion coverage** when you only have a still, screenshot, frame grab, or generated image
* **Bring a character portrait to life** while keeping identity as consistent as possible
* **Create atmospheric establishing shots** from landscape, architecture, or environment stills
* **Generate short cinematic inserts** for edits that need motion but do not need a fully new scene
#### Getting Started
**Step 1: Add Your Image**
1. Open Chat Video Pro inside Premiere Pro
2. Click **Studio**
3. In the **Production** department, click **Motion Director**
4. Add a still image from **Upload**, **Recents**, or **Frame Capture**
Motion Director only accepts still images.
Source
Best Use
Upload
A still from your computer, a client frame, a design export, or external key art
Recents
A frame you generated in Cinematic Lab, Multi-Cam, Relight Scene, or chat
Frame Capture
The current frame from your Premiere timeline, useful when matching an existing edit
The better the source still, the better the motion. Motion Director can add camera movement, foreground drift, parallax, and ambient dynamics — but it cannot rescue a weak frame. Start with a clear subject, readable depth, and enough room for the camera to move.
{% hint style="info" %}
**Best source images have depth.** Motion Director performs best when the image has foreground, subject, and background layers. A flat product cutout on a blank background gives the camera less to work with than a product on a table with props behind it.
{% endhint %}
**Step 2: Configure the Camera Move**
Once your image is loaded, Motion Director opens the configure screen.
The screen has three main areas:
* **Your Image** — preview of the source still
* **Camera Movement** — click this card to choose the movement preset
* **Composer** — optional action text, reference elements, voice input, duration, speed, resolution, audio, and Generate
The UI is intentionally simple, but the underlying generation is more structured than a normal prompt. Motion Director uses Kling 3.0 Pro Image-to-Video, which looks at your image and identifies the main subject. It then builds a movement-specific prompt that separates **camera movement** from **subject action**, adds identity preservation language, and chooses a movement-appropriate negative prompt.
That matters because image-to-video models can easily do the wrong thing: morph a face, move a parked car, create extra characters, or turn a dolly out into a fake zoom. Motion Director's presets are designed to prevent those mistakes as much as possible.
***
#### Choosing a Camera Movement
Click the **Camera Movement** card to open the movement picker.
Camera moves out through a foreground object, leaving the scene
Exit beats, scene endings, stylized transitions, mystery reveals
Action Run
Camera tracks a running subject with energetic forward motion
Chase beats, sports, action trailers, high-energy social clips
Vehicle Chase
Camera follows alongside or behind a moving vehicle
Car commercials, pursuit scenes, travel B-roll, automotive hero shots
Movement Guide
**Dolly In**
Use **Dolly In** when the shot should get more intimate or more important over time. It works best with portraits, products, and centered subjects.
**Good action text:**
> “subtle wind moves through her hair, city lights flicker behind her”
**Avoid:**
* Asking the subject to run toward camera
* Very crowded backgrounds with no foreground separation
* Source images where the subject is already extremely close to camera
**Dolly Out**
Use **Dolly Out** when the environment matters. The subject naturally gets smaller as more of the surrounding world is revealed.
**Good action text:**
> “dust hangs in the air as the ruined city is revealed behind him”
**Avoid:**
* Tight portraits with no surrounding scene to reveal
* Asking the model to keep the subject the exact same size — that turns into a Vertigo-style move
**Orbit**
Use **Orbit** for beauty shots and subject reveals. It creates parallax by moving around the subject, which makes stills feel more three-dimensional.
**Good action text:**
> “light catches the edge of the jacket as the camera arcs around”
**Avoid:**
* Flat front-facing images with no side information
* Full 360-degree expectations from a short clip. Motion Director's built-in orbit is a safer 180-degree arc.
**Vertigo**
Use **Vertigo** for psychological tension. It is the Hitchcock dolly zoom: the camera pulls while the lens zooms, so the background appears to compress behind the subject.
**Good action text:**
> “the room seems to close in as the character realizes what happened”
**Avoid:**
* Calling it “warp,” “bubble,” or “fisheye.” Those words can push the model into distorted lens effects instead of a grounded dolly zoom.
**Tracking Shot**
Use **Tracking Shot** when the subject is moving and the camera should follow alongside. This is the one movement where subject motion is expected, so your action text matters more.
**Good action text:**
> “the subject walks left to right through a busy market, coat moving in the wind”
The workflow exposes direction controls for supported tracking shots, such as **Left to Right** or **Right to Left**.
**Avoid:**
* Static subjects with no clear reason to move
* Contradictory action like “standing completely still” while choosing a tracking move
**Crane Up**
Use **Crane Up** when you want the camera to rise and reveal scope. It is slower, more formal, and more cinematic than a simple zoom.
**Good action text:**
> “morning fog rolls across the field as the camera rises above the character”
**Avoid:**
* Cluttered ceilings or indoor scenes without vertical room
* Short durations. Crane Up usually needs 10 seconds to breathe.
**Aerial Reveal**
Use **Aerial Reveal** for geography: landscapes, city blocks, houses, events, campuses, beaches, roads, and anything where the world is the subject.
**Good action text:**
> “the drone pulls back to reveal the entire coastline at golden hour”
**Avoid:**
* Close-up faces
* Images with no visible environment beyond the subject
**FPV Drone**
Use **FPV Drone** when energy matters more than precision. It works best with environments that have clear paths, depth, and obstacles to fly past.
**Good action text:**
> “the camera banks past neon signs and dives toward the street below”
**Avoid:**
* Portraits with no background depth
* Scenes where identity preservation is more important than motion energy
**Handheld**
Use **Handheld** when you want a human camera operator feel: small breathing movement, micro-jitters, documentary realism.
**Good action text:**
> “the subject glances slightly off-camera as dust moves through the light”
**Avoid:**
* Product shots that need perfect stability
* Camera moves that should be mathematically clean, like dolly or crane
**Crash Zoom**
Use **Crash Zoom** when the shot needs a sudden snap-in toward the subject. It creates immediate visual impact.
**Good action text:**
> "the subject's eyes widen as the camera snaps closer"
**Avoid:**
* Slow, meditative scenes where a gentle push-in would read better
* Wide landscapes with no clear focal subject
**Snorricam**
Use **Snorricam** when the subject should stay locked in frame while the world moves around them. The effect feels like the camera is mounted to the person.
**Good action text:**
> "the city blurs past as the character stumbles forward, disoriented"
**Avoid:**
* Static portraits where the subject should not move
* Product shots that need a stable hero frame
**Earth Zoom Out**
Use **Earth Zoom Out** for epic scale reveals — from a close subject to a planet-wide view.
**Good action text:**
> "the camera pulls back from the hiker to reveal the entire mountain range below"
**Avoid:**
* Indoor close-ups with no visible geography to reveal
* Shots where the subject must stay large in frame
**Luxury Tabletop Turn**
Use **Luxury Tabletop Turn** for premium product and tabletop hero shots. The camera arcs slowly around an object on a surface.
**Good action text:**
> "specular highlights glide across the watch face as the camera turns"
**Avoid:**
* Full-body character shots where the subject is not on a surface
* Fast action scenes that need energetic camera work
**Through Object In**
Use **Through Object In** when the camera should push through a foreground object to enter the scene — a window, doorway, leaf, or lens element.
**Good action text:**
> "the camera passes through rain on the glass into the warm interior"
**Avoid:**
* Flat images with no foreground layer to pass through
* Shots where the subject is already fully visible with nothing to reveal
**Through Object Out**
Use **Through Object Out** as the reverse — the camera exits the scene by moving through a foreground object.
**Good action text:**
> "the camera pulls back through the doorway, leaving the room behind"
**Avoid:**
* Open landscapes with no foreground exit point
* Shots where you need the subject to remain visible at the end
**Action Run**
Use **Action Run** when a subject is running and the camera should track with forward energy.
**Good action text:**
> "the runner sprints down the alley, coat flying, camera keeping pace"
**Avoid:**
* Static subjects with no reason to run
* Slow, contemplative scenes
**Vehicle Chase**
Use **Vehicle Chase** when a vehicle is the moving subject and the camera follows alongside or from behind.
**Good action text:**
> "the car accelerates through the tunnel, headlights cutting the darkness"
**Avoid:**
* Parked vehicles with no motion
* Indoor product shots where vehicle movement does not apply
***
#### Composer Controls
Below the movement selector, Motion Director gives you a compact composer.
**Optional Action Text**
The text box is for **scene dynamics**, not the main camera move. The camera move is already chosen by the preset. Use this box to describe what happens while the camera moves.
Good action text:
* “wind blows through her hair.”
* “rain streaks down the window”
* “dust floats in the sunbeam.”
* “The subject slowly turns toward the camera.”
* “City lights flicker in the distance.”
* “fabric moves gently in the breeze.”
Weak action text:
* “Make it cinematic.”
* “add motion”
* “cool camera movement”
* “Fix the face.”
* “Change the outfit.”
{% hint style="warning" %}
**Do not fight the camera move.** If you choose Dolly In, do not write “camera pulls away.” If you choose Tracking Shot, describe what the subject is doing and which direction they move. Contradictory action text is one of the fastest ways to get unstable results.
{% endhint %}
#### AI Optimize ✨
Tap the sparkle (✨) button next to the action description to get an AI-improved version of your prompt. The optimizer understands the selected movement and any reference image you have loaded — so the result is grounded in your specific shot rather than generic. Choose **Replace** to apply it, **Regenerate** to try again, or **Close** to keep your original.
#### Voice Input
You can use the microphone button to dictate the action text. This is often faster than typing because you can describe movement naturally:
> “slow fog in the background, her hair moves slightly, the neon sign flickers once as the camera pushes in.”
**Reference Elements**
Motion Director lets you attach up to **6 reference elements**. These are extra images that help Kling keep a character, product, prop, outfit, or identity more consistent while the camera moves.
You can attach elements from:
* **Upload**
* **Recents**
* **Saved Elements**
* **Frame Capture** from Premiere
{% hint style="info" %}
**Pro tip — use elements for character consistency.** If you are animating a portrait or recurring character, create or attach one clean element of the subject: face visible, outfit visible, no heavy motion blur. This gives the generation a stronger identity anchor than the source image alone.
{% endhint %}
#### Duration, Speed, Resolution, and Audio
Control
Options
How to Think About It
Duration
3–15 seconds
Shorter for punchy moves, longer for complex moves that need space
Speed
Slow, Medium, Fast
Controls movement intensity, not clip duration
Direction
Shown for supported moves
Currently useful for Tracking Shot-style lateral movement
Resolution
1080p or 4K
Use 1080p for tests, 4K for keepers
Audio
Off / On
Optional native audio from Kling; leave off if you plan to design sound in Premiere
**Speed Recommendations**
Speed
Best For
Slow
Luxury, drama, beauty, real estate, emotional beats
Medium
Default for most shots
Fast
Action, FPV, energetic social edits, dramatic reveals
Use **Slow** when you care about realism. Use **Fast** when energy matters more than perfect preservation.
***
#### Working With Your Results
After generation finishes, Motion Director opens a single-video preview.
You can:
* **Play the video** and inspect the motion
* **Regenerate** if the move is close but not right
* **Back** to adjust movement, duration, speed, resolution, audio, or action text
* **Done** to add the finished video to chat
When you click **Done**, the video is saved as a normal Chat Video Pro video message with metadata for the selected movement, intensity, model, and duration. From the chat you can review it, save it, reuse it from Recents, or bring it into Premiere.
{% hint style="warning" %}
**Generation can take several minutes.** Kling 3.0 Pro is slower than image generation, especially at longer durations or 4K. Keep the panel open while the job runs. If the ETA passes, the workflow may show a high-volume message while it continues polling.
{% endhint %}
***
#### Pro Tips
**Create the still first, then direct the camera**
Do not try to solve composition and motion at the same time. Use Cinematic Lab to create the still, then Motion Director to animate it. You will get better results by separating the photography decision from the camera-movement decision.
**Use elements when the subject matters**
If the hero is a person, product, mascot, car, or character, attach a reference element. A clean subject element can preserve face, outfit, product shape, or brand details far better than the source frame alone. This is the most important pro move for character consistency.
**Keep action text subordinate to the camera**
The movement preset is the director. Your text box is the background actor. Write things that complement the move: wind, rain, flicker, fabric motion, slight head turn, dust, smoke, light pulses. Avoid competing movement like “camera pulls back” inside a Dolly In preset.
**Give wide moves room**
Aerial Reveal, Crane Up, and Dolly Out need space around the subject. If your source image is tightly cropped, the model has nothing to reveal and may invent unstable background details. For those moves, start with a wider still or generate a wider frame in Cinematic Lab first.
**Use Handheld to make AI stills feel less synthetic**
Even a very subtle Handheld pass can make a generated still feel like footage captured by a real operator. It is not flashy, but for documentary, interviews, gritty promos, or emotional portraits, it can be more believable than a dramatic orbit.
**Use Orbit for product and character beauty, not complex action**
Orbit wants a subject that can be studied. A perfume bottle, sneaker, car, character portrait, costume, sculpture, or hero prop works. A crowded street, complex battle scene, or fast action frame usually does not.
**Plan for the 16:9 output**
If the source image is vertical, leave extra horizontal room or generate a 16:9 version first. A vertical portrait animated into 16:9 can lose composition if the subject is too tight or centered without background.
**Regenerate one variable at a time**
If a result is close, do not change movement, speed, duration, resolution, and action text all at once. Change one thing, regenerate, compare. Otherwise you will not know what improved or broke the shot.
***
#### Example Ideas
**Cinematic Lab → Motion Director**
Generate a still in Cinematic Lab:
> “A lone astronaut standing on the edge of a Martian crater at sunrise, dust in the air, Earth visible as a tiny blue point above the horizon.”
Then animate it in Motion Director:
Setting
Choice
Movement
Aerial Reveal
Duration
10–15s
Speed
Slow
Action Text
“dust drifts across the crater as the camera rises”
**Product Hero Shot**
Source image:
> A luxury watch on black velvet with brushed metal reflections.
| Setting | Choice |
| ----------- | ---------------------------------------------------------------- |
| Movement | **Dolly In** or **Orbit** |
| Duration | **5–10s** |
| Speed | **Slow** |
| Action Text | “specular highlights move across the metal as the camera glides” |
| Element | Clean product reference, if available |
**Real Estate Reveal**
Source image:
> Exterior home shot at golden hour.
| Setting | Choice |
| ----------- | --------------------------------------------------------- |
| Movement | **Aerial Reveal** or **Crane Up** |
| Duration | **10–15s** |
| Speed | **Slow / Medium** |
| Action Text | “warm light moves across the roofline, trees sway subtly” |
**Character Trailer Beat**
Source image:
> A masked villain standing in a smoke-filled hallway.
| Setting | Choice |
| ----------- | ---------------------------------------------------------------- |
| Movement | **Dolly In** or **Vertigo** |
| Duration | **5s** |
| Speed | **Medium** |
| Action Text | “smoke curls around the shoulders, overhead lights flicker once” |
| Element | Clean character face / costume reference |
***
#### Troubleshooting
**The subject's face or outfit changes**\
Attach a reference element of the subject and keep your action text simple. Avoid asking for wardrobe, age, expression, or identity changes if you want preservation.
**The subject starts walking, driving, or moving when I only wanted camera motion**\
Remove action text and use a camera-only movement such as Dolly In, Dolly Out, Orbit, Crane Up, or Aerial Reveal. Motion Director adds stricter stillness language when no action text is provided.
**The camera move feels too subtle**\
Increase **Speed** from Slow to Medium or Fast. If the move still feels weak, choose a movement with stronger physical transformation, such as Dolly In, Crane Up, or Aerial Reveal.
**The camera move feels too chaotic**\
Lower **Speed**, shorten the action text, and use 1080p test generations before committing to 4K. FPV Drone and Handheld are intentionally less stable than Dolly / Crane / Orbit.
**The result looks like a zoom instead of a real camera move**\
Use a source image with more depth: foreground, subject, and background. Dolly and Orbit need visual layers to show parallax.
**The output framing surprised me**\
Motion Director currently outputs 16:9. If you started with a vertical or square image, try generating a wider source still first or choose a move that does not require as much horizontal room.
**The video takes a long time**\
Kling 3.0 Pro can take several minutes, especially at longer durations or 4K. Keep the panel open while it runs.
**Generation fails immediately**\
Check Settings and confirm your FAL API key is configured and funded. Motion Director needs FAL access to run Kling generation.
**Text or logos in the image distort during motion**\
This is common in image-to-video. Use slower movement, shorter duration, and a cleaner source image. If the text or logo is mission-critical, keep the camera movement very subtle.
***
**Next:** Learn about AI Transitions to bridge two still frames into a moving shot, or return to Cinematic Lab to create a stronger source image before animating.
# AI Transitions
AI Transitions takes a start frame and an end frame, then generates the motion between them. Instead of adding a normal dissolve in Premiere, you can create a real AI bridge: a product reveal, whip pan, smoke reveal, flame burn, focus pull, time passage, flying camera move, or seamless transformation where the first image physically becomes the second.
{% hint style="info" %}
**AI Transitions works best when both frames are intentional.** Treat the start and end frames like the first and last frames of a shot. If either frame is weak, mismatched, or poorly composed, the transition has to solve too many problems at once.
{% endhint %}
#### When to Use AI Transitions
Use AI Transitions when you have two stills that should feel connected by motion.
* **Bridge two Cinematic Lab frames** into a moving shot
* **Turn before/after images** into a satisfying reveal
* **Create product reveals** from silhouette, packaging, closed box, or detail frames
* **Connect two locations** with a wipe, whip pan, or flying camera move
* **Show time passage** from day to night, clean to messy, old to new, empty to full
* **Make a transformation feel physical** instead of using a flat cross-dissolve
* **Create stylized edit punctuation** for trailers, ads, music videos, gaming edits, explainers, and social clips
AI Transitions is strongest when the relationship between the two frames is clear. The model should understand what is changing, what is staying the same, and what kind of motion should explain the change.
{% hint style="warning" %}
**Both frames must have matching aspect ratios.** If the start frame is 16:9, the end frame must also be 16:9. If the start frame is 9:16, the end frame must also be 9:16. The workflow checks this before generation and blocks mismatched pairs.
{% endhint %}
#### Getting Started
**Step 1: Add a Start Frame**
The **Start Frame** is the first frame of the generated shot.
You can add it from:
* **Upload**
* **Recents**
* **Frame Capture** from the current Premiere playhead
Use a frame that clearly establishes the subject, environment, and visual direction. The model analyzes this frame first and uses it as the starting anchor.
**Step 2: Add an End Frame**
The **End Frame** is the final target of the generated shot.
Use a frame that matches the start frame's aspect ratio and has a believable relationship to it. The best end frame answers the question: "Where should this shot land?"
Good pairs:
| Start Frame | End Frame | Why It Works |
| -------------------- | --------------------------- | ----------------------------- |
| Closed product box | Product fully revealed | Clear transformation target |
| Messy room | Clean room | Strong before/after structure |
| Daytime city street | Same street at night | Natural time-passage logic |
| Character in shadow | Character fully lit | Dramatic reveal |
| Wide empty landscape | Final hero subject in scene | Scene reveal / flying camera |
Weak pairs:
* Different subjects with no clear relationship
* Different aspect ratios
* One frame close-up and one frame extreme wide, unless the selected style explains that move
* Frames with conflicting camera angles and no transition style that can justify the change
* Two images that both contain lots of text, logos, or detailed UI
***
#### Step 3: Choose a Transition Style
Click the center **Transition** card to open the style picker.
Each style is more than a visual label. It carries a prompt structure: how to lock the first frame, how to describe the sequence, what the camera should do, and what artifacts to avoid.
Style
Best For
Seamless Morph
Faces, costume changes, object transformations, age progression
Fast edits, vlogs, action cuts, invisible scene changes
Logo Transform
Brand intros, logo reveals, identity transitions, creative commercial openers
Portal
Dimensional portal transitions; sub-styles include Mystic Sparks, Cartoon Dimension, Sci-Fi Oval, Reality Window, and Tech Rift
Glass Reflection Flip
Reflective surface flips that reveal the next scene through glass or mirror
Film Burn Reveal
Film-burn wipe that exposes the end frame through organic light leak
Transition Style Guide
**Seamless Morph**
Use **Seamless Morph** when the subject in the first frame physically becomes the subject in the second frame.
Best for:
* Face morphs
* Costume changes
* Product state changes
* Object transformations
* Age progression
Avoid using it when the two images are totally different scenes. A morph needs shared structure.
**Before & After**
Use **Before & After** when the two frames are the same space or subject in different conditions.
Best for:
* Dirty to clean
* Broken to repaired
* Unedited to edited
* Empty to furnished
* Unstyled to styled
This style works best when objects line up spatially. If the camera angle changes too much between frames, the wipe may feel unstable.
**Product Reveal**
Use **Product Reveal** when the transition should feel like an ad: light sweep, fog, rim light, hero reveal.
Best for:
* E-commerce videos
* Luxury products
* Packaging reveals
* Brand intros
* Beauty shots
For best results, keep the product in roughly the same screen position in both frames.
**Time Passage**
Use **Time Passage** when the scene changes because time has moved forward.
Best for:
* Day to night
* Construction progress
* City timelapse
* Weather shift
* Room filling with people
Keep the camera position as similar as possible between start and end frames. The more stable the geography, the better the timelapse reads.
**Scene Wipe**
Use **Scene Wipe** when a foreground object should pass close to camera and hide the cut.
Best for:
* Dream sequences
* Memory flashes
* Location changes
* Invisible cuts
* Story transitions
It works best when you can imagine an object crossing the lens: a person, wall, car, pillar, tree, flag, door, or dark shape.
**Flying Cam**
Use **Flying Cam** when the transition should feel like a fast camera move through space.
Best for:
* Travel reveals
* Real estate
* Action sequences
* Establishing shots
* Environments with visible depth
Flying Cam needs room to travel. It struggles with flat portraits or tight product shots with no background depth.
**Smoke Reveal**
Use **Smoke Reveal** when you want atmosphere to hide and reveal the subject.
Best for:
* Character intros
* Product reveals
* Music videos
* Dramatic moments
* Horror, mystery, or fantasy tones
Smoke works best with high contrast and lighting direction. If both frames are flatly lit, the reveal may feel like a dissolve.
**Flame Transition**
Use **Flame Transition** when the edit needs energy and impact.
Best for:
* Action edits
* Gaming content
* Music videos
* Sports
* High-drama trailer moments
Avoid it for subtle corporate, documentary, or luxury scenes unless the brand can handle the intensity.
**Focus Pull**
Use **Focus Pull** when the transition is about attention shifting from one plane to another.
Best for:
* Revealing a detail
* Dialogue moments
* Narrative discovery
* Object to person, foreground to background, clue to reaction
The two frames should feel like they could exist in the same shot with different focus planes.
**Liquid/Melt**
Use **Liquid/Melt** for surreal physical transformations.
Best for:
* Creative morphs
* Album art motion
* Abstract product campaigns
* Beauty / fashion surrealism
* Artistic social content
Liquid/Melt is intentionally stylized. It is not the best choice for realistic continuity.
**Whip Pan**
Use **Whip Pan** when speed should hide the change.
Best for:
* Vlog cuts
* Fast-paced social edits
* Action beats
* Music videos
* Invisible location jumps
Whip Pan works especially well when both frames have strong horizontal composition.
**Logo Transform**
Use **Logo Transform** when the transition should use a logo or brand mark as the visual mechanism of change.
Best for:
* Brand intros and openers
* Logo reveals and identity moments
* Corporate and commercial punctuation
* Creative transitions anchored to a brand asset
Logo Transform works best when at least one of the frames features a logo, wordmark, or recognizable brand shape.
**Portal**
Use **Portal** when the transition should open or close through a dimensional gateway. Choose a sub-style to match the tone:
* **Mystic Sparks** — magical energy and particle bursts.
* **Cartoon Dimension** — playful, stylized portal animation.
* **Sci-Fi Oval** — sleek futuristic gateway.
* **Reality Window** — a window into another scene.
* **Tech Rift** — digital glitch or tech-driven tear.
Best for fantasy, sci-fi, gaming, music videos, and stylized brand moments where the transition itself is part of the story.
**Glass Reflection Flip**
Use **Glass Reflection Flip** when the next scene should appear through a reflective surface — a mirror, window, or polished glass panel flipping to reveal the destination frame.
Best for:
* Elegant product or fashion transitions.
* Interior-to-exterior reveals.
* Premium commercial punctuation.
* Shots where reflection and refraction add visual interest.
**Film Burn Reveal**
Use **Film Burn Reveal** when the transition should feel like organic film stock burning away to expose the end frame.
Best for:
* Vintage or analog aesthetics.
* Music videos and trailers.
* Nostalgic brand moments.
* Edits where a warm light-leak wipe fits the visual language.
***
#### Composer Controls
Below the three-frame setup is the composer row.
**Notes**
Use the text box to add transition direction. Keep it short and physical.
Good notes:
* "Make the wipe travel left to right."
* "Use golden light particles."
* "Keep the camera locked off."
* "Make the smoke reveal slowly and elegantly."
* "Make the final frame feel like a luxury product hero shot."
* "Use fast directional blur, not a dissolve."
Weak notes:
* "Make it cool."
* "cinematic"
* "better"
* "smooth transition"
* "do everything"
The selected style already contains the main transition logic. Your notes should steer emphasis, not rewrite the whole shot.
#### AI Optimize ✨
Tap the sparkle (✨) button next to the notes field to get an AI-improved version of your transition notes. The optimizer understands your selected transition style and the start/end frames — so the suggestion is specific to your actual editorial moment rather than generic. Choose **Replace** to apply it, **Regenerate** to try again, or **Close** to keep your original.
**Voice Input**
You can use the microphone button to dictate notes. This is useful when you want to describe the editorial beat quickly:
> "Start locked on the messy room, then let the wipe reveal the clean room slowly from left to right, like a satisfying before-after reel."
**Duration**
AI Transitions supports the full **3-15 second** duration range.
Duration
Best For
3-4s
Whip Pan, punchy social edits, fast scene changes
5-6s
Product Reveal, Smoke Reveal, Focus Pull, Flame, Before & After
8-10s
Seamless Morph, Flying Cam, Time Passage
10-15s
Slow environmental reveals or complex transformations
Most styles have a recommended duration marker. Use it as a starting point before experimenting.
**Audio**
The **Audio** toggle asks Kling to generate native audio with the transition.
Leave it **Off** when:
* You are cutting to music
* You plan to design sound in Premiere
* You want full control over SFX
Turn it **On** when:
* The transition has obvious physical sound potential, like flame, whip pan, flying camera, or smoke
* You want a quick temp sound bed for review
***
#### Working With Results
After generation finishes, AI Transitions opens a single-video preview.
You can:
* **Play the result**
* **Back** to adjust style, duration, resolution, audio, or notes
* **Regenerate** by returning to setup and running again
* **Done** to add the transition video to chat
When you click **Done**, the video is saved as a normal Chat Video Pro video message. The message stores the transition style, model, duration, audio setting, and detected aspect ratio metadata.
{% hint style="warning" %}
**Do not close the panel while generation is running.** Kling O3 Pro can take several minutes. Long transitions can run 15-20 minutes. If the job times out or your connection drops, AI Transitions may show a **Resume** option so you can reconnect to the existing job instead of starting over.
{% endhint %}
***
#### Pro Tips
**Design the first and last frame before choosing the transition**
The style is the bridge, not the destination. Start by making sure both frames are strong. If the end frame is unclear, generate a better one in Cinematic Lab or Multi-Cam before trying to transition.
**Keep the subject relationship obvious**
The best transitions have a clear "same thing, different state" or "same camera energy, new location" relationship. If the model cannot tell what connects the frames, it may create a mushy hallucinated middle.
**Match aspect ratio and visual language**
The workflow enforces aspect ratio, but it cannot enforce art direction. If one frame is photorealistic and the other is stylized, the middle will usually wobble. Match lighting, realism, focal length, and framing before generation.
**Use frame capture for timeline-native transitions**
Park the Premiere playhead on the last frame of one shot and capture it as the Start Frame. Then capture or generate the destination frame. This is the fastest way to make AI Transitions feel connected to your actual edit.
**Use Recents as your transition tray**
Generate stills in Cinematic Lab, Multi-Cam, or Relight Scene, click Done, then pull them from Recents inside AI Transitions. This avoids export/import friction and keeps the whole chain inside Chat Video Pro.
**Give complex transitions more time**
Time Passage, Flying Cam, and Seamless Morph often need 8-10 seconds. Whip Pan and Flame can work faster. If the transition feels rushed, increase duration before changing the prompt.
**Use 1080p to choose the idea, 4K to finish**
Do not burn 4K generations while deciding whether Smoke Reveal or Product Reveal is the right direction. Find the style at 1080p, then run the keeper at 4K.
**Use notes to specify direction**
If direction matters, say it: "left to right," "camera pushes through the doorway," "wipe travels upward," "smoke parts from the center." Directional clarity reduces random motion.
**Avoid asking for a hard cut**
The value of AI Transitions is the in-between. If you want a true cut, do it in Premiere. Use AI Transitions when the middle motion matters.
Example Ideas
**Product Box to Hero Reveal**
| Setting | Choice |
| ----------- | ------------------------------------------------------------ |
| Start Frame | Product box closed in dark studio |
| End Frame | Product fully revealed in hero lighting |
| Style | **Product Reveal** |
| Duration | **6s** |
| Notes | "slow rim light sweep, premium product ad, no repositioning" |
**Messy Room to Clean Room**
| Setting | Choice |
| ----------- | -------------------------------------------- |
| Start Frame | Messy room |
| End Frame | Same room cleaned and styled |
| Style | **Before & After** |
| Duration | **6s** |
| Notes | "wipe left to right, keep furniture aligned" |
**Day to Night City Timelapse**
| Setting | Choice |
| ----------- | ------------------------------------------------------ |
| Start Frame | City street at golden hour |
| End Frame | Same street at night with neon and traffic |
| Style | **Time Passage** |
| Duration | **10s** |
| Notes | "locked camera, light trails, buildings remain stable" |
**Character Intro from Smoke**
| Setting | Choice |
| ----------- | ----------------------------------------------------- |
| Start Frame | Subject mostly hidden in fog |
| End Frame | Subject fully visible in dramatic light |
| Style | **Smoke Reveal** |
| Duration | **6s** |
| Notes | "volumetric smoke parts from the center, no dissolve" |
**Fast Location Jump**
| Setting | Choice |
| ----------- | ------------------------------------------- |
| Start Frame | Creator in one location |
| End Frame | Creator in a second location |
| Style | **Whip Pan** or **Scene Wipe** |
| Duration | **3-5s** |
| Notes | "fast horizontal motion blur hides the cut" |
***
#### Troubleshooting
**The second frame is rejected**\
The aspect ratio does not match the start frame. Regenerate or crop one frame so both are the same ratio.
**The transition looks like a dissolve**\
Use a more physical style, such as Flame, Smoke Reveal, Scene Wipe, Liquid/Melt, or Whip Pan. Add notes like "not a dissolve" and "the change is physical."
**The subject changes too much in the middle**\
Use frames with stronger structural similarity, or choose Before & After / Product Reveal instead of Seamless Morph. Make sure both frames clearly show the same subject.
**The camera drifts when it should stay locked**\
Use notes like "locked camera," "fixed tripod," or "no camera movement." Time Passage, Before & After, Focus Pull, and Seamless Morph usually want a stable camera.
**The background warps or melts**\
The frames may be too different, or the transition duration may be too short. Try a longer duration and a style that explains the change more clearly.
**The job takes a long time**\
Kling O3 Pro transitions can take several minutes, and longer clips can run much longer. Keep the panel open. If the workflow offers **Resume**, use it to reconnect to the existing job.
**Generation fails immediately**\
Check Settings and confirm your FAL API key is configured. AI Transitions needs FAL access to create the Kling job.
**The result is almost right but the timing feels wrong**\
Keep the same frames and style, then adjust duration only. Do not change every variable at once.
***
**Next:** Learn about Relight Scene to change lighting and mood on a still or clip, or return to Cinematic Lab to create stronger start and end frames before transitioning.
# Avatar Studio
Avatar Studio generates lip-synced talking-head presenter videos from a photo or script — no camera, no shoot, no scheduling required.
{% hint style="info" %}
**Avatar Studio generates video, not still images.** Each generation produces a finished, lip-synced clip ready to drop directly onto your Premiere Pro timeline. There is no frame-by-frame editing inside Avatar Studio — it is a one-take generation workflow.
{% endhint %}
**When to Use Avatar Studio**
Avatar Studio is the right tool any time you need a presenter on screen without booking a shoot.
* **UGC-style social ads** — produce authentic-looking spokesperson clips at scale for Instagram Reels, TikTok, and YouTube Shorts
* **Product walkthroughs and explainers** — walk viewers through features with a talking presenter in the frame without scheduling talent
* **Title-card and overlay presenters** — generate a transparent-background spokesperson and composite them over any footage in Premiere Pro
* **Rapid ad variant testing** — change the script, change the voice, generate again without re-booking anyone
* **Any project where you need a talking head in five minutes** — if you have a script and an avatar, you have a clip
***
**Getting Started**
**Step 1: Open Avatar Studio**
1. Open Chat Video Pro (Window → Extensions → Chat Video Pro)
2. Click the **Studio** button in the sidebar to open the Launchpad
3. In the **Production** department, click the **Avatar Studio** card
***
**Step 2: Choose Your Avatar**
Avatar Studio gives you two ways to pick who appears on screen.
**Preset Avatars**
Browse the built-in gallery and select any presenter. Each card shows the presenter's name and a preview thumbnail. Preset avatars support the full range of background controls — including transparent output — and include a background-capable indicator when those controls are available.
**Your Photo**
Select **Your photo** and upload a portrait (JPEG, PNG, or WebP; at least 256 × 256 px; up to 15 MB). The face should be clearly visible, well-lit, forward-facing, and unobstructed. Your photo mode has its own resolution range and always renders on the default background.
{% hint style="info" %}
Toggle between Preset and Your photo at any time before generating. Background color and transparent settings are only active in preset avatar mode — switching to Your photo automatically resets to the default background.
{% endhint %}
***
**Step 3: Set Up Your Voice and Script**
Two paths determine what the avatar says and how it sounds.
**Script + Voice**
Type or paste your script into the text box — up to **5,000 characters** per generation. Pick a voice from the curated voice chips, or use the search field to browse additional voices. Click the play icon on any voice chip to hear a preview before committing.
{% hint style="info" %}
Write the way you speak, not the way you write. Short sentences, natural pauses (add a comma or period where you would breathe), and conversational language produce more natural-sounding delivery. Read the script aloud before generating — if it sounds stiff when you read it, it will sound stiff from the avatar. For longer content, split the script into multiple generations and assemble the clips on the Premiere timeline.
{% endhint %}
**My Audio**
Upload an audio file (MP3, WAV, M4A, AAC, or OGG — up to 120 seconds). Avatar Studio drives the lip-sync directly from your recording. No voice selection is required. Your exact inflections, pacing, and pauses are preserved.
Use **My Audio** when the delivery must match a specific performance — a particular accent, emotional tone, or reading cadence that a preset voice cannot replicate.
***
**Step 4: Output Settings**
Before generating, configure the output to match your platform and use case.
**Aspect Ratio**
Six options: **16:9** (landscape/widescreen), **9:16** (vertical — Stories, Reels, Shorts), **4:5** (portrait feed), **5:4**, **1:1** (square), and **Auto** (the engine selects based on the avatar source). Pick the ratio that matches your target platform before generating — it cannot be changed after.
Platform
Ratio
TikTok / Instagram Reels / YouTube Shorts
9:16
Instagram feed / Facebook
4:5 or 1:1
YouTube / website / horizontal ad
16:9
**Resolution**
Available resolutions depend on your avatar source:
* **Preset avatar** — 720p, 1080p, or 4K. For most social content, 1080p is the sweet spot.
* **Your photo** — 360p, 480p, 540p, 720p, or 1080p. Higher resolutions produce sharper output but take longer to render.
**Background** *(preset avatars with background support only)*
Background controls appear for preset avatars that include background compositing support. Three options:
* **Default** — the avatar renders on its natural built-in scene.
* **Color** — choose any solid color. Quick-access swatches include green screen, white, black, studio blue, and charcoal; a full color picker lets you dial in any hex value.
* **Transparent** — the avatar renders with a transparent background, encoded as ProRes 4444. Premiere Pro reads the alpha channel natively — no keying or matte step required. Use this for title overlays, lower-thirds, or compositing the presenter over your own footage.
**Talking Style** *(Your photo only)*
* **Stable** — consistent, measured facial movement. Best for product explainers and professional content.
* **Expressive** — more animated facial expression. Better for social ads and casual-tone content.
**Generate SRT Captions**
When this toggle is on, Avatar Studio downloads a synchronized SRT captions file alongside the video. Import it into Premiere Pro as a subtitle track to caption the presenter with no manual transcription.
***
**Step 5: Add to Your Timeline**
Click **Generate**. A progress indicator shows while the video renders — typically about a minute. When the preview appears, watch it to confirm the result, then click **Add to chat** to attach it as a chat message asset.
From there, drag the clip directly onto your Premiere Pro timeline the same way you would any other generated video in Chat Video Pro. If you generated a transparent-background ProRes 4444 clip, place it on a track above your main footage — the alpha channel is preserved automatically, with no keying step required.
***
**Pro Tips**
**Use a clear, well-lit, forward-facing photo for best results**
The face should fill a good portion of the frame with no sunglasses, masks, or heavy shadows. Avatar Studio processes and crops the background automatically — the photo doesn't need a clean backdrop, but the face does.
**Transparent background lets you composite over anything in Premiere**
Choose a background-capable preset avatar, set the background to Transparent, and generate. Because the result is ProRes 4444, Premiere Pro reads the alpha channel natively — no keying step required. Drop the clip onto a track above your footage and it sits cleanly over any background you have.
**Use My Audio when the performance matters**
If you've already recorded a line with a specific delivery, pacing, or accent that a preset voice can't replicate, My Audio preserves it exactly. It's also the fastest path when you have a polished voice-over recording already in your project.
**Preview voices before committing**
The play icon on each voice chip plays a short sample before you select. Running through a few options takes less than a minute and is faster than committing, generating, and discovering the wrong voice afterward.
**Generate multiple variations to find the best take**
Change the voice, talking style, or script and generate again — each result is independent. Collect the clips you want in chat, then compare them on the timeline before picking your final version.
**Split long scripts into segments**
Avatar Studio supports up to 5,000 characters per generation. For longer content, break the script at natural pause points, generate each segment separately, and assemble the clips in order on the Premiere timeline.
***
**Troubleshooting**
**The Generate button is disabled**\
Avatar Studio requires an API key configured in Settings. If it is missing or invalid, generation will not start. Go to Settings → API Keys, confirm the key is entered, and check that it is funded.
**Lip-sync looks off or the face is stiff**\
For Your photo mode, use a higher-resolution source with a clear, unobstructed face. For Script + Voice, try shortening sentences and adding punctuation at natural pause points — it improves cadence and can noticeably smooth the delivery.
**The transparent background option is grayed out**\
Transparent output is only available for preset avatars that include background compositing support. Switch to a background-capable preset avatar to unlock the option. Your photo mode does not support transparent backgrounds.
**The output uses a background I didn't expect**\
Background controls only appear for preset avatars with background support. If you chose Your photo or a preset without background capability, the avatar always renders on its default scene.
**The SRT captions file didn't appear**\
The Generate SRT Captions toggle must be on before clicking Generate — the captions file generates alongside the video and cannot be added retroactively. Turn on the toggle and generate again.
**The video has the wrong aspect ratio**\
Aspect ratio cannot be changed after generation. If the result is the wrong shape, adjust the ratio setting and generate again.
**Generation failed or timed out**\
Try again — video generation can occasionally fail under high load. If failures continue, check your API key status in Settings and confirm the key is funded.
***
**Next:** Layer a transparent-background Avatar Studio presenter over any scene in Premiere Pro, or use Relight Scene to match the lighting of an AI-generated background to your presenter clip.
# Rotoscoping
Professional-grade background removal and object isolation using Meta's SAM 3
{% embed url="" %}
Rotoscope is the Studio workflow for isolating a subject from its background. Use it when you want to keep a person, product, prop, animal, or object and turn the rest of the frame transparent for compositing.
Chat Video Pro uses **SAM 3 Rotoscoping** to identify the subject, track it through the clip, and create a transparent video output that can be placed over new footage, graphics, generated backgrounds, or motion design inside Premiere Pro.
### What This Tool Is For
Rotoscope is for **keeping the selected subject** and removing everything else.
Use it for:
* Talking-head background removal.
* Product cutouts.
* Isolating a dancer, athlete, actor, or presenter.
* Pulling a foreground object out of a shot.
* Creating transparent overlays for edits, thumbnails, trailers, tutorials, and ads.
* Preparing a subject to place over AI-generated backgrounds or design elements.
Do not use Rotoscope when your goal is to remove a distracting object from the scene while keeping the original background. That is an object-erasing/inpainting problem, so use Erase Objects instead.
{% hint style="info" %}
**Rotoscope answers: "What should stay?"** Erase Objects answers: "What should disappear?"
{% endhint %}
### When To Use It
Use Rotoscope when the subject is worth separating from the scene:
Goal
Why Rotoscope helps
Put a presenter over a new background
Creates a transparent subject layer for compositing.
Isolate a product demo
Keeps the product motion while removing the original environment.
Build a graphic overlay
Lets you layer a person or object above text, titles, or b-roll.
Create social cutouts
Makes subjects reusable across vertical edits, thumbnails, and promo assets.
Combine live footage with AI scenes
Extracts the live subject so it can sit over generated backgrounds.
Choose another Studio workflow when:
Goal
Better workflow
Remove a person or object and keep the same background
Erase Objects
Change part of a clip with a prompt
Reshoot
Add rain, fire, smoke, or style
Add Effects
Improve lighting or mood
Relight Scene
Increase resolution after editing
Upscale
***
### Studio Path
The fastest route is through Studio.
1. Open **Studio**.
2. Choose **Rotoscope** from the Post-Production department.
3. Load a video from upload, Recents, or your Premiere timeline.
4. Choose a selection method: **Text**, **Box**, or **Point**.
5. Click **Track Frame** to preview the mask.
6. If the mask is right, click **Track Entire Video**.
7. Click **Remove Background** to create the transparent result.
8. Send the result to your chat/library and bring it back into Premiere.
Studio opens the video editor directly in SAM 3 Rotoscoping mode, so you can start selecting the subject immediately.
***
### Classic/Editor Path
You can also use Rotoscope from the classic video editor path:
1. Import or generate a video.
2. Click **Edit** on the video thumbnail.
3. Choose **SAM 3 Rotoscoping** if it is not already selected.
4. Select the subject.
5. Track the frame, track the full video, and remove the background.
This path is useful when you are already working from a chat result. For new post-production work, Studio is cleaner because it starts in the right tool.
***
### Controls And Constraints
| Control | What it does |
| ------------------ | ------------------------------------------------------------------------------------------- |
| Text | Selects the subject using a description like `the person`, `the red car`, or `the product`. |
| Box | Lets you draw a box around the subject for more precise control. |
| Point | Lets you click points on the subject. Shift-click marks areas to exclude. |
| Track Frame | Tests the selection on the current frame before processing the full clip. |
| Track Entire Video | Tracks the approved subject through the whole video. |
| Remove Background | Converts the tracked result into a transparent video for compositing. |
Current constraints:
| Constraint | Detail |
| ----------------------- | ------------------------------------------------------------------------------------------------- |
| Model | SAM 3 video segmentation. |
| Recommended clip length | Short clips are faster and easier to verify. Under 30 seconds is a good working target. |
| Maximum clip length | SAM 3 supports longer clips, up to about 5 minutes, but processing time scales with length. |
| Source quality | Clear, well-lit, high-contrast subjects track better. |
| Output | Transparent video for Premiere compositing, with a panel-friendly preview generated for playback. |
| Reference images | Not used. Rotoscope selects from the video itself. |
***
### Selection Methods
#### Text Selection
Use Text when the subject is obvious and easy to describe.
Good text prompts:
```
the person
```
```
the dancer in the center
```
```
the red car
```
```
the product on the table
```
Text is best for clean scenes with one clear subject. It is usually the fastest way to start.
#### Box Selection
Use Box when the scene has multiple subjects, busy backgrounds, or a subject that text might misunderstand.
Best practices:
* Draw the box tightly around the subject.
* Include the full subject, not just the face or center.
* Avoid including large background areas.
* Use Shift-drag to mark exclusion areas when needed.
* Start on the first frame when using spatial selection.
Box selection is often the safest choice for production work because it gives the model a strong spatial hint.
#### Point Selection
Use Point when you need a quick selection or want to guide SAM 3 toward a specific object.
Best practices:
* Click near the center of the subject.
* Add multiple include points for larger subjects.
* Shift-click areas you want excluded.
* Use Box instead if the subject has a complex shape or overlaps other objects.
* Start on the first frame when using spatial selection.
Point selection is fast, but it can be less stable than a good box on difficult footage.
***
### The Rotoscope Workflow
#### 1. Select The Subject
Decide what should remain visible. Rotoscope works best when you think in terms of the final composited layer:
* Keep the presenter.
* Keep the product.
* Keep the car.
* Keep the dancer.
* Keep the foreground prop.
Avoid vague selections like `foreground` or `everything important`. Name the actual subject.
#### 2. Track Frame
Track Frame creates a preview mask before you spend time processing the full clip.
Look for:
* The correct subject is selected.
* Edges are close enough for the intended use.
* Important limbs, hair, products, or props are included.
* Background areas are not being included by mistake.
* Multiple subjects are not being merged unless you want them together.
If the preview is wrong, adjust the selection method and track the frame again.
#### 3. Track Entire Video
Once the frame preview is good, Track Entire Video processes the clip. Chat Video Pro runs the preview video and mask-data work together, so the result can be used for the final transparent export.
Longer clips take longer. If you are testing a difficult subject, process a short segment first before committing to a long one.
#### 4. Remove Background
After tracking completes, click **Remove Background**. Chat Video Pro uses the cached mask data to create a transparent result. The current pipeline uses the original video plus SAM 3 mask data to produce a cleaner alpha result, with a preview that can still play inside the panel.
The final result is designed for Premiere compositing, not just browser preview.
***
### Best Practices
#### Start With The Cleanest Frame
For Box and Point selections, start at the first frame. The video workflow expects spatial selections to begin at the start of the clip, and you will get more predictable tracking when the subject is visible right away.
If the subject is not visible on frame one, trim the clip so the first frame is a strong selection frame.
#### Keep Clips Short While Testing
Rotoscoping is easier to diagnose in short pieces. For a long shot, test a 5-10 second segment first. Once you know the selection works, process the longer section or split the scene into manageable parts.
#### Use Contrast To Your Advantage
SAM 3 does better when the subject is visually distinct from the background. High-contrast clothing, clean lighting, visible outlines, and stable framing all help.
Hard cases include:
* Dark clothing on a dark background.
* Fast motion blur.
* Thin hair against complex detail.
* Transparent objects.
* Multiple overlapping people.
* Subjects leaving and re-entering frame.
#### Choose The Right Selection Method
Use Text for obvious subjects. Use Box for precision. Use Point for quick correction or simple object picks. If one method fails, switch methods rather than repeating the same bad selection.
#### Composite Intentionally
A transparent subject is only half the shot. Once you bring it into Premiere:
* Place it above the new background.
* Match scale and position.
* Add color correction so the subject belongs in the scene.
* Add shadows, blur, or grain when needed.
* Use feathering or additional masks in Premiere for edge cleanup if the shot demands it.
***
### Examples
#### Talking Head On A New Background
* Source: Presenter on a plain or messy background.
* Selection: Box around the presenter or text prompt `the person`.
* Result: Presenter on transparent background, ready for a branded graphic or generated scene.
#### Product Cutout
* Source: Product demo video.
* Selection: Text prompt `the product` or a tight box around the item.
* Result: Product motion isolated for ads, thumbnails, landing pages, or overlays.
#### Dance Layer
* Source: Dancer footage.
* Selection: Box around the dancer on the first frame.
* Result: Dancer isolated over typography, generated backgrounds, or music-video visuals.
#### Foreground Object Extraction
* Source: Clip with a car, prop, animal, or object in front of the camera.
* Selection: Text prompt or box around the target.
* Result: Object separated for compositing or reuse in another sequence.
***
### Troubleshooting
#### Track Frame is disabled
Make a selection first. Text mode needs a prompt. Box mode needs a drawn box. Point mode needs at least one point.
#### Box or Point selection gives an error
Scrub to the beginning of the clip and select on the first frame. Spatial selections are most reliable from frame zero.
#### The wrong subject is selected
Use a more specific text prompt or switch to Box selection. If there are multiple people or objects, describe position, color, or role:
```
the person in the blue jacket
```
```
the car in the foreground
```
#### The mask is right on the first frame but drifts later
Process a shorter segment, use a clearer first-frame selection, or split the shot around difficult moments. Drift often happens when the subject turns, gets blocked, leaves frame, or overlaps another subject.
#### The background is not fully removed
Return to the selection step and tighten the box, add exclusion points, or use a simpler subject description. If the footage has low contrast, try a cleaner source or shorter segment.
#### The transparent result does not preview like a normal MP4
Transparent video formats are different from standard playback formats. Chat Video Pro creates a panel-friendly preview for viewing, while the real transparent output is meant for editing/compositing in Premiere.
***
### Links To Related Studio Pages
* [Studio](/features/studio) - Learn how Studio workflows are organized.
* [Erase Objects](/features/studio/object-eraser-tool) - Remove unwanted people or objects instead of keeping them.
* [Reshoot ](/features/studio/reshoot)- Regenerate a targeted part of a clip.
* [Add Effects ](/features/studio/kling-vfx)- Add atmosphere, VFX, or style to existing footage.
* [Relight Scene](/features/studio/relight-scene) - Improve lighting before or after isolating a subject.
* [Upscale](/features/studio/video-upscaling) - Improve resolution after rotoscoping or generation.
***
**Next:** If you want to remove something from the shot while keeping the original background, use Erase Objects.
# Erase Objects
Remove unwanted objects, people, logos, or any elements from your videos using AI. VOID intelligently fills in the background where the object was, making it disappear completely.
Erase Objects is the Studio workflow for removing something unwanted from a short video while keeping the rest of the shot. Use it for bystanders, logos, signs, boom mics, cups, clutter, background distractions, or small objects that pull attention away from the edit.
Chat Video Pro uses **VOID** to identify what should disappear, remove it, and fill the empty area with a generated background that matches the surrounding footage.
***
### What This Tool Is For
Erase Objects is for **removing the selected object** and preserving the rest of the scene.
Use it for:
* Removing a bystander from b-roll.
* Cleaning up a product shot.
* Removing a logo, sign, timestamp, or watermark where you have the rights to do so.
* Removing a mic boom, tripod, cable, cup, tag, or small production mistake.
* Cleaning short social clips before delivery.
* Testing whether an object can be removed before spending time on manual cleanup.
Do not use Erase Objects when you want to keep a subject and remove the entire background. That is a rotoscoping task, so use Rotoscope instead.
{% hint style="info" %}
**Erase Objects answers: "What should disappear?"** Rotoscope answers: "What should stay?"
{% endhint %}
\*\*\*
### When To Use It
Use Erase Objects when the unwanted item is visible, describable, and surrounded by enough background for the model to rebuild the area.
Goal
Why Erase Objects helps
Remove a passerby
Cleans up b-roll without reshooting.
Remove a logo or sign
Makes footage less distracting or more brand-safe.
Remove gear from frame
Cleans up boom mics, stands, cables, and accidental equipment.
Remove clutter
Simplifies interviews, desks, product shots, and social clips.
Remove a small foreground object
Lets the surrounding texture fill in naturally.
Choose another Studio workflow when:
Goal
Better workflow
Keep a subject and remove the whole background
Rotoscope
Replace or change part of a clip with a prompt
Reshoot
Add VFX, weather, or style
Add Effects
Change lighting or mood
Relight Scene
Improve resolution after cleanup
Upscale
***
### Studio Path
The fastest route is through Studio.
1. Open **Studio**.
2. Choose **Erase Objects** from the Post-Production department.
3. Load a video from upload, Recents, or your Premiere timeline.
4. Choose **Text**, **Point**, or **Draw** selection.
5. Describe or mark what should be removed.
6. Choose an output format if needed.
7. Click **Erase Object**.
8. Review the result and send it back to your chat/library.
Studio opens the video editor directly in VOID mode, so you do not need to select the model manually.
***
### Classic/Editor Path
You can also use Erase Objects from the classic video editor path:
1. Import or generate a short video.
2. Click **Edit** on the video thumbnail.
3. Choose **VOID** from the model selector.
4. Select what to erase using Text, Point, or Draw mode.
5. Choose output format.
6. Click **Erase Object**.
This path is useful when you are already working from a chat result. For new cleanup work, Studio is the better starting point because it opens the correct tool immediately.
***
### Controls And Constraints
Control
What it does
Text
Removes an object based on a written description.
Point
Lets you click the object directly. Click marks erase points; Shift-click marks protect points.
Draw
Lets you paint a region around the object to define the area to erase.
Output Format
Chooses the result codec/container: H.264, H.265, ProRes, or VP9.
Erase Object
Starts the VOID cleanup job.
Current constraints:
Constraint
Detail
Model
VOID (by Netflix).
Frame rate
VOID expects 20-30 fps. Chat Video Pro prepares incompatible clips automatically when possible.
Best source
Stable, well-lit footage with a clear object to remove.
Selection methods
Text, Point, or Draw (region). All three modes are supported.
Audio
The workflow preserves audio when processing.
Reference images
Not used. Erase Objects works from the source video and your selection.
#### Output Formats
Format
Best for
H.264 (MP4)
Most compatible, best default for review and general editing.
H.265 (MP4)
Smaller files when your workflow supports H.265.
ProRes (MOV)
Higher-quality editing handoff, larger files.
VP9 (WebM)
Web-oriented output.
If you are unsure, use H.264. If you are sending the result into a heavier finishing workflow and file size is not a concern, ProRes can be a better handoff format.
***
### Selection Methods
#### Text Selection
Use Text when the object is easy to describe.
Good prompts:
```
the person in the red shirt
```
```
the logo in the bottom right corner
```
```
the microphone boom at the top of the frame
```
```
the water bottle on the table
```
Text mode works best when there are not many similar objects in the shot. If there are multiple people, signs, cups, or cars, include location and visual details.
Less effective:
```
person
```
```
the thing
```
```
remove it
```
#### Point Selection
Use Point when the object is hard to describe or when you want more direct control.
Point mode behavior:
* Click on the object to mark it for erasure.
* Add multiple erase points for larger or irregular objects.
* Shift-click areas you want to protect.
* Use protect points when the model is removing nearby detail you want to keep.
* Clear all points and try again if the selection gets confusing.
#### Draw Selection
Use Draw when you want to paint a region around the object rather than clicking discrete points. This is useful for irregular shapes, larger areas, or objects that are difficult to click precisely.
Draw mode behavior:
* Paint over the object to define the area to erase.
* Larger brush strokes cover irregular or spread-out objects more easily than individual click points.
* Use Draw when Point mode misses edges or the object has an unusual outline.
***
### Best Practices
#### Trim To The Exact Problem
Trim aggressively. Do not send extra lead-in or tail frames unless they are needed for the cleanup. VOID works best on shorter, focused clips where the object to remove is clearly visible throughout.
Good uses:
* A shot where a person briefly crosses the background.
* A product close-up with a tag or hand to remove.
* A short interview moment with a cup or cable in frame.
Bad uses:
* A long clip with a distraction for only a couple of seconds — trim to just that section.
* A long handheld shot with the object moving behind multiple subjects.
* A wide shot with a tiny object that is barely visible.
#### Remove One Clear Thing At A Time
VOID works best when the instruction is focused. If you need to remove several unrelated objects, do separate passes or start with the most distracting object.
Better:
```
the coffee cup on the desk
```
Riskier:
```
the cup, the cables, the logo, the chair, and the person in the background
```
#### Give The Model Background To Rebuild
Object removal is easier when the area behind the object is predictable: wall, floor, sky, desk, road, grass, or a repeated texture.
Harder cases include:
* Faces behind the object.
* Text or signs behind the object.
* Complex reflections.
* Fast camera motion.
* Objects crossing detailed hands, hair, or patterned clothing.
* Large objects covering a big portion of frame.
#### Use Protect Points
If Point mode starts erasing nearby details, Shift-click areas you want to protect. This is useful when removing an object close to a face, product edge, logo you want to keep, or another person.
#### Review The Fill, Not Just The Removal
A result can remove the object but still look wrong if the generated fill is smeared, warped, or inconsistent. Watch the entire cleaned clip, especially the frames before and after the object moves.
***
### Examples
#### Clean B-Roll
* Source: Street b-roll with a passerby in the background.
* Selection: `the person walking in the background`.
* Result: Cleaner shot for a brand, travel, or corporate edit.
#### Remove Production Gear
* Source: Interview clip with a boom mic entering the top of frame.
* Selection: `the microphone boom at the top of the frame`.
* Result: Usable interview moment without sending the clip to a manual cleanup pass.
#### Product Shot Cleanup
* Source: Product video with a sticker, tag, hand, or stray object.
* Selection: Point mode on the item or prompt `the tag on the product`.
* Result: Cleaner product shot for ads, landing pages, or social content.
#### Remove A Logo Or Sign
* Source: Background sign or corner logo.
* Selection: `the logo in the bottom right corner`.
* Result: Less distracting footage, assuming you have the right to remove it.
***
### Troubleshooting
#### The model removes the wrong object
Use a more specific prompt or switch to Point mode. Include location, color, size, or relationship to the frame:
```
the person on the far left
```
```
the white cup in front of the laptop
```
#### The object is only partly removed
Use Point mode and add more erase points on the remaining pieces. For text mode, name the full object more clearly:
```
the entire microphone boom and its shadow
```
#### The background fill looks unnatural
Try a shorter clip, a more stable section, or a shot where the hidden background is simpler. VOID has to invent what was behind the object, so complex backgrounds are harder.
#### Nearby details are disappearing
Use protect points with Shift-click in Point mode. Mark the object with erase points, then mark nearby details that should stay.
#### The clip needs frame-rate conversion
VOID expects 20-30 fps constant-frame-rate video. Chat Video Pro prepares the clip automatically when possible. If preparation fails, try exporting a short H.264 MP4 from Premiere at 24, 25, or 30 fps and use that as the source.
***
### Links To Related Studio Pages
* [Studio](/features/studio) - Learn how Studio workflows are organized.
* [Rotoscope](/features/studio/sam-3-rotoscoping) - Keep a subject and remove the background.
* [Reshoot](/features/studio/reshoot) - Make broader targeted changes to a shot.
* [Add Effects](/features/studio/kling-vfx) - Add atmosphere, VFX, or character swaps.
* [Relight Scene](/features/studio/relight-scene) - Change lighting or mood.
* [Upscale ](/features/studio/video-upscaling)- Improve resolution after cleanup.
***
**Next:** If you want to keep a subject and remove everything around it, use Rotoscope.
# Add Effects
Add visual effects to your videos without After Effects. Add rain, fire, lighting changes, weather effects, and more. Supports character swapping with reference images.
{% embed url="" %}
Instead of generating a brand-new clip from scratch, Add Effects sends your source video into a video-to-video model that edits the existing motion and composition. The default path is **Kling O3 VFX**; you can also use **Google Omni Flash**, **Grok**, **Wan 2.7**, or **Luma Ray 3.2 Edit** depending on the effect. Luma Ray 3.2 Edit brings a 4-way divergence dial — Default, Adhere, Balanced, Reimagine — that sets how far the restyle strays from the source (note it generates no audio). **Effect layers** let you stack effects on the same clip. The goal is to keep the shot recognizable while changing the layer of reality on top of it.
***
### What This Tool Is For
Add Effects is best for broad visual transformations that affect a whole shot or a clearly described part of a shot.
Push the shot toward cyberpunk, noir, vintage film, dream sequence, horror, or commercial polish.
Character or object swap
Use reference images or saved Elements to replace a person, creature, prop, or visual motif.
Add Effects is not a full compositing replacement. It is strongest when you want a believable transformation over a short section of footage, not frame-perfect manual control over every particle, mask, or layer.
{% hint style="info" %}
**Think of Add Effects as AI art direction for existing footage.** You are not rebuilding the shot from nothing. You are telling the model how the current shot should feel, what should change, and what must remain stable.
{% endhint %}
### When To Use It
Use Add Effects when you want to:
* Test a VFX idea before committing to a heavier edit.
* Give ordinary footage a more cinematic lighting or weather treatment.
* Create a social clip that needs a visual hook in a few seconds.
* Match a shot to a mood board, reference image, or generated still.
* Replace a character or object with an attached reference.
* Build several creative variations from the same clip and choose the best one in context.
Choose a different Studio tool when the task is more specific:
Goal
Better workflow
Change lighting direction or mood on an image or short video
Relight Scene
Remove a person or object from a short clip
Erase Objects
Replace or modify a specific section of a shot
Reshoot
Remove the background and isolate the subject
Rotoscope
Increase resolution after generation
Upscale
Create a new shot from a still image
Motion Director
***
### Studio Path
The fastest route is through Studio.
1. Open **Studio**.
2. Choose **Add Effects** from the Post-Production department.
3. Load a video from your computer, Recents, or your Premiere timeline.
4. Add an edit prompt that describes the effect or transformation.
5. Optional: add up to four reference images or Elements.
6. Choose **Pro** or **Standard** quality.
7. Generate the effect.
8. Use the before/after view to compare the result, then send it back to your chat or library.
Studio automatically opens the video editor in the correct Kling VFX mode, so you do not need to manually pick the model unless you intentionally switch tools.
***
### Classic/Editor Path
You can also reach the same tool from the classic video editor flow.
1. Import or generate a video in Chat Video Pro.
2. Click **Edit** on the video thumbnail.
3. Open the model selector.
4. Choose **Kling VFX**.
5. Write your effect prompt and add references if needed.
6. Generate and compare the result.
This path is still useful when you are already working from a chat result or an older library item. For new post-production work, Studio is the cleaner starting point because it loads the right tool immediately.
***
### Controls And Constraints
Control
What it does
Edit Prompt
Describes the effect, style change, character swap, or visual transformation.
Quality
Pro uses Kling O3 Pro VFX for higher-quality results. Standard is the faster/lighter option.
Reference Images
Adds up to four images that can guide style, characters, props, or objects.
Elements
Lets you pull saved character/object Elements into the reference slots.
@Image1, @Image2 tags
Tells the model which attached reference image to use.
Before/After
Compares the generated result against the original source clip.
Current constraints:
Constraint
Detail
Video duration
Source clips must be at least 3 seconds. Clips longer than about 10 seconds can be trimmed to a selected segment inside the editor.
Recommended duration
5-8 seconds is usually the sweet spot for quality and consistency.
Reference images
Up to 4 images. Use one strong reference when possible; use multiple only when each one has a clear job.
Prompt length
Keep the prompt focused. The model responds better to clear direction than long lists of unrelated effects.
Audio
Add Effects keeps the original audio automatically.
Resolution
Very large or non-compliant videos may be converted before generation so Kling can process them reliably.
***
#### AI Optimize ✨
Tap the sparkle (✨) button next to the edit prompt to get an AI-improved version of your effect description. The optimizer understands the effect prompt and the current video frame — so the suggestion targets what's actually in your clip rather than offering a generic VFX phrase. Choose **Replace** to apply it, **Regenerate** to try again, or **Close** to keep your original.
***
### Two Ways To Prompt
#### Video Only
Use this when the effect is simple enough to describe in words.
Good for:
* Rain, snow, fog, smoke, sparks, fire, dust, or atmosphere.
* Day-to-night or golden-hour transformations.
* Broad style changes like noir, horror, vintage, or cyberpunk.
* Quick experiments where you do not need an exact visual target.
Example:
{% code overflow="wrap" %}
```
Add heavy rain throughout the scene, wet pavement reflections, water droplets on the lens, and a moody cinematic night atmosphere. Keep the original camera movement and subject motion.
```
{% endcode %}
#### Video + Reference Image
Use this when you care about the exact look. The reference image gives Kling a visual target instead of making it infer everything from text.
Good for:
* Matching a generated still from Cinematic Lab.
* Turning a frame into a specific lighting or weather design first, then applying that look to the video.
* Character or costume swaps.
* Complex style transformations that are hard to describe.
Example:
{% code overflow="wrap" %}
```
Transform the video to match the lighting and atmosphere of @Image1. Keep the same composition and camera movement, but apply the neon night look, wet street reflections, and cool blue-magenta color contrast from the reference.
```
{% endcode %}
***
### Best Practices
#### Describe The Effect, Not A New Scene
Kling VFX is editing the uploaded video. The best prompts explain what should change in the existing shot.
Good:
{% code overflow="wrap" %}
```
Add drifting fog through the background, subtle beams of light from camera left, and a cooler haunted-house atmosphere. Preserve the original actor, camera move, and scene layout.
```
{% endcode %}
Less effective:
{% code overflow="wrap" %}
```
A person walking through a foggy haunted house.
```
{% endcode %}
The second prompt sounds like text-to-video. It does not clearly tell the model what to preserve from your source footage.
#### Give Each Reference A Job
If you attach references, explain what each image controls.
{% code overflow="wrap" %}
```
Use @Image1 as the lighting and color reference. Use @Image2 as the creature design reference. Add the creature emerging from the background shadows while preserving the original camera movement.
```
{% endcode %}
If you only say "make it look like the image," the model has to guess whether the image represents color, lighting, subject identity, costume, composition, or all of the above.
#### Start With A Strong Clip
Add Effects works best when the original video has:
* Clear subject/background separation.
* Stable enough motion for the model to understand the shot.
* A short duration with one main action.
* Enough visual detail for the effect to attach to surfaces, light, and movement.
Shaky, dark, heavily compressed, or visually chaotic footage gives the model less structure to preserve.
#### Use [Cinematic Lab](/features/studio/cinematic-lab) Or [Relight Scene](/features/studio/relight-scene) As Prep Tools
For complex looks, create a target still first.
1. Capture a frame from the video.
2. Use Cinematic Lab or Relight Scene to design the target look.
3. Bring that image into Add Effects as `@Image1`.
4. Prompt Kling to match the lighting, atmosphere, color, or effect from the reference.
This workflow is much stronger than trying to describe a highly specific visual style in one sentence.
#### Preserve What Matters
If the shot has important details, say so directly:
{% code overflow="wrap" %}
```
Keep the actor's face, wardrobe, body motion, and camera move unchanged. Add a supernatural blue glow from the doorway and light fog rolling along the floor.
```
{% endcode %}
For character swaps:
{% code overflow="wrap" %}
```
Replace the person in the video with the character shown in @Image1. Preserve the same pose, timing, walking motion, and camera framing. Keep the background unchanged.
```
{% endcode %}
***
### Examples
#### Rainy Night Street
{% code overflow="wrap" %}
```
Turn the scene into a rainy night street. Add heavy rain, wet pavement reflections, visible droplets in the foreground, darker sky, and cinematic streetlight glow. Keep the original subject motion and camera movement.
```
{% endcode %}
#### Golden Hour Commercial Look
{% code overflow="wrap" %}
```
Apply warm golden-hour lighting across the scene, soft sun flare from camera right, gentle haze, polished commercial contrast, and natural skin tones. Preserve the original scene layout and motion.
```
{% endcode %}
#### Horror Atmosphere
{% code overflow="wrap" %}
```
Add low drifting fog, colder moonlit shadows, subtle flickering practical lights, and a tense horror atmosphere. Do not change the actor or camera move.
```
{% endcode %}
#### Cyberpunk Transformation With Reference
{% code overflow="wrap" %}
```
Use @Image1 as the lighting and color reference. Transform the scene into a neon cyberpunk night environment with blue-magenta reflections, wet surfaces, and glowing signage. Preserve the original subject and camera motion.
```
{% endcode %}
#### Character Swap
{% code overflow="wrap" %}
```
Replace the person in the video with the character from @Image1. Match the character's outfit, silhouette, and face as closely as possible while preserving the original movement, timing, and background.
```
{% endcode %}
#### Subtle Product Polish
{% code overflow="wrap" %}
```
Improve the shot with subtle cinematic lighting, cleaner contrast, soft atmospheric depth, and a premium product-ad look. Keep the product shape, label, camera framing, and motion unchanged.
```
{% endcode %}
***
### Troubleshooting
#### The effect barely changed
Make the prompt more specific. Name the effect, where it should appear, and how intense it should be.
Try:
{% code overflow="wrap" %}
```
Add dense fog in the background and low rolling mist across the floor, strongest near the doorway, with cool blue backlight and visible atmosphere around the subject.
```
{% endcode %}
#### The result changed too much
Add preservation language:
{% code overflow="wrap" %}
```
Preserve the original subject identity, wardrobe, camera movement, composition, and timing. Only change the lighting and atmosphere.
```
{% endcode %}
#### The reference image did not guide the result
Make sure the reference image is attached and mentioned by tag, such as `@Image1`. Then explain what the image is for:
{% code overflow="wrap" %}
```
Use @Image1 only as the lighting and color reference. Do not copy the subject or composition from the image.
```
{% endcode %}
#### The video is too short
Kling VFX needs at least 3 seconds of video. Use a longer section from the timeline or choose another workflow.
#### The video is too long
The editor can trim longer videos to a selected segment for Kling VFX. Pick the 5-10 seconds where the effect matters most, generate that section, then place the result back into your Premiere edit.
#### The video needs conversion
Some videos need normalization before Kling can process them. If Chat Video Pro asks to convert the clip, allow the conversion unless you specifically need to preserve the original file format outside the tool. The generated result will still be used as a normal video asset.
#### The effect works, but the style is inconsistent
Use a target image. Capture a frame, create the desired look in Cinematic Lab or Relight Scene, then use that image as `@Image1` in Add Effects.
***
### Links To Related Studio Pages
* [Studio](/features/studio) - Learn how Studio workflows are organized.
* [Cinematic Lab](/features/studio/cinematic-lab) - Create high-quality reference stills for VFX direction.
* [Relight Scene](/features/studio/relight-scene) - Design lighting and mood changes before applying them to video.
* [Motion Director](/features/studio/motion-director) - Turn a still image into a new moving shot.
* [AI Transitions ](/features/studio/ai-transitions)- Morph between two frames with directorial transition prompts.
* [Multi-Cam](/features/studio/multi-cam) - Generate alternate angles from an image or video.
* [Erase Objects](/features/studio/object-eraser-tool) - Remove unwanted people or objects from short clips.
* [Reshoot](/features/studio/reshoot) - Make more targeted changes with LTX Retake.
* [Rotoscope](/features/studio/sam-3-rotoscoping) - Isolate subjects and remove backgrounds.
* [Upscale](/features/studio/video-upscaling) - Improve resolution after generating or editing footage.
***
**Next:** If you want a more controlled lighting change instead of a broader VFX transformation, use Relight Scene.
# Reshoot
Change specific areas or sections of your video using LTX Reshoot. Remove objects, add elements, or modify parts of a scene while keeping the rest of your footage unchanged. Perfect for targeted edits
{% embed url="" %}
Reshoot is the Studio workflow for changing a selected part of a video without starting over. Use it when a shot is close, but one moment needs a new action, a different object, cleaner timing, replacement audio, or a more specific beat.
Chat Video Pro uses **LTX Retake** to regenerate the selected segment while using the surrounding clip as context. Think of it like asking for a new take inside an existing shot.
### What This Tool Is For
Reshoot is for **targeted retakes**. It is not a general style filter, rotoscope tool, or long-form editor.
Use it to:
* Regenerate one short moment in a clip.
* Change what a person or object does.
* Add, remove, or alter a scene element.
* Replace a short section of audio.
* Fix a small scene issue without reworking the full video.
* Try a different action while preserving the surrounding shot context.
{% hint style="info" %}
**Reshoot answers: "What should happen differently in this selected segment?"** The clearer that answer is, the better the result.
{% endhint %}
### When To Use It
Use Reshoot when the main problem is a specific moment, action, or element:
Goal
Why Reshoot helps
Change a short action
Regenerates the selected moment with a new prompt.
Replace a scene detail
Lets you modify an object or part of the shot.
Fix a bad beat
Gives you a new version of a specific time range.
Regenerate audio and video together
AV Sync mode creates a synced replacement take.
Regenerate only audio
Audio Only mode can change sound while keeping the picture.
Choose another Studio workflow when:
Goal
Better workflow
Remove an object from a clip
Erase Objects
Keep a subject and remove the background
Rotoscope
Add rain, fire, atmosphere, or style
Add Effects
Change lighting or mood
Relight Scene
Increase resolution after editing
Upscale
Reshoot is strongest when the request is local and concrete. If you want the entire clip to become cyberpunk, snowy, noir, or golden hour, use Add Effects or Relight Scene.
***
### Studio Path
The fastest route is through Studio.
1. Open **Studio**.
2. Choose **Reshoot** from the Post-Production department.
3. Load a video from upload, Recents, or your Premiere timeline.
4. Drag the timeline handles to select the segment you want to regenerate.
5. Describe the new content or change.
6. Choose the retake mode: **AV Sync**, **Video Only**, or **Audio Only**.
7. Click **Reshoot Segment**.
8. Compare the result and click **Done** when you want to send it back to your chat/library.
Studio opens the video editor directly in LTX Reshoot mode. You do not need to manually choose the model.
***
### Classic/Editor Path
You can also use Reshoot from the classic video editor path:
1. Import or generate a video.
2. Click **Edit** on the video thumbnail.
3. Choose **LTX Reshoot** from the model selector.
4. Select the segment.
5. Write the reshoot prompt.
6. Choose retake mode.
7. Generate and review.
This path is useful when you are already working from a chat result. For new targeted fixes, Studio is cleaner because it opens the correct tool immediately.
***
### Controls And Constraints
Control
What it does
Timeline handles
Select the time range that will be regenerated.
Prompt
Describes what should happen in the replacement segment.
AV Sync
Regenerates both video and audio. This is the default mode.
Video Only
Regenerates video while keeping the original audio.
Audio Only
Regenerates audio while keeping the original video.
Reshoot Segment
Starts the LTX Retake job for the selected range.
Current constraints:
Constraint
Detail
Model
LTX Video 2.0 Retake.
Segment duration
Minimum 2 seconds, maximum 20 seconds.
Default selection
Starts at 0 seconds and uses up to 5 seconds, capped by clip length.
Best working range
5-10 seconds is usually easiest to control.
Source compatibility
Chat Video Pro prepares incompatible video formats for Fal when possible.
Reference images
Not used. Reshoot works from the video and prompt.
***
### Retake Modes
#### AV Sync
Use **AV Sync** when the new moment needs picture and sound to work together.
Best for:
* A person saying or reacting differently.
* A new action that should have matching audio.
* A scene beat where motion and sound both matter.
* Generating a fresh take of the selected moment.
#### Video Only
Use **Video Only** when the picture should change but the original audio should stay.
Best for:
* Removing or modifying a visual detail while preserving dialogue or music.
* Changing a movement while keeping production sound.
* Fixing the look of a segment without touching audio timing.
#### Audio Only
Use **Audio Only** when the visual is fine but the sound needs a different take.
Best for:
* Replacing a sound moment.
* Trying a cleaner audio beat.
* Keeping the original video while exploring generated audio changes.
***
### Writing Reshoot Prompts
Reshoot prompts should describe the replacement moment as directly as possible.
Good prompts:
```
The person waves goodbye at the camera.
```
```
Remove the small sign on the wall and keep the background natural.
```
```
The car changes from blue to red while keeping the same movement.
```
```
Replace the old poster with a clean blank wall.
```
```
The person looks down at the product and smiles.
```
Less effective prompts:
```
Make it better.
```
```
Make it cinematic.
```
```
A person walking through a city.
```
The weak prompts either do not say what should change or sound like a request for a brand-new clip instead of a targeted retake.
#### AI Optimize ✨
Tap the sparkle (✨) button next to the reshoot prompt to get an AI-improved version of your retake description. The optimizer understands the retake description and the source clip — so the result is a concrete, targeted replacement rather than a vague rewrite. Choose **Replace** to apply it, **Regenerate** to try again, or **Close** to keep your original.
***
### Best Practices
#### Select The Smallest Useful Segment
Do not reshoot the whole clip unless the whole clip needs a new take. Shorter segments are easier to control and easier to judge.
Good segment choices:
* The 3 seconds where the person reacts.
* The 5 seconds where an object should change.
* The one beat where the audio should be replaced.
Riskier segment choices:
* A full 20-second section with several unrelated actions.
* A segment where the subject enters and exits frame multiple times.
* A broad scene change that should really be a new generation or VFX pass.
#### Say What Changes And What Stays
If parts of the shot should remain stable, include that in the prompt.
{% code overflow="wrap" %}
```
Change the blue car to red while keeping the same street, camera movement, and traffic timing.
```
{% endcode %}
```
Remove the cup from the desk. Keep the person, laptop, and background unchanged.
```
This helps LTX focus the retake instead of reimagining the entire selected moment.
#### Use Reshoot For Scene Logic, Not Broad Style
Reshoot can add or change scene elements, but it is not the best tool for broad atmospheric transformations. If the note is "make this night with rain and neon," use Add Effects. If the note is "change this cup into a phone for the next 4 seconds," use Reshoot.
#### Start With Video Only When Audio Matters
If you are editing footage with dialogue, music, or production sound you already like, try **Video Only** first. It keeps the audio stable while changing the picture.
Use **AV Sync** when the new video action needs new sound to make sense.
#### Compare Before Committing
Watch the transition into and out of the selected segment. A good reshoot is not just a good generated middle; it should also cut back into the original clip cleanly.
***
### Examples
#### Change A Product Moment
* Segment: 4 seconds where a product is held up.
* Mode: Video Only.
* Prompt: `The person holds the product closer to camera and turns it slightly toward the light.`
* Result: A more useful product beat while preserving the original audio.
#### Remove A Small Scene Element
* Segment: 3 seconds where an unwanted poster is visible.
* Mode: Video Only.
* Prompt: `Remove the poster on the back wall and replace it with a plain wall. Keep the person and camera movement unchanged.`
* Result: Cleaner background without running a full object-erasing workflow.
#### New Reaction Take
* Segment: 5 seconds on a character reaction.
* Mode: AV Sync.
* Prompt: `The person notices something off camera, smiles, and says a short surprised reaction.`
* Result: A new action/audio beat for the same moment.
#### Audio Retake
* Segment: 4 seconds with distracting sound.
* Mode: Audio Only.
* Prompt: `Replace the noisy audio with clean room tone and subtle footsteps.`
* Result: Same video with a cleaner generated audio moment.
***
### Troubleshooting
#### The result changed too much
Shorten the segment and add preservation language:
{% code overflow="wrap" %}
```
Keep the person, camera movement, background, and timing unchanged. Only remove the cup from the desk.
```
{% endcode %}
#### The requested change did not happen
Make the prompt more concrete. Name the object, action, location, and what should replace it.
```
Replace the blue sign on the left wall with a blank white wall.
```
#### The wrong part of the clip changed
Adjust the timeline handles. Reshoot only knows the selected segment, so make sure the selection starts before the change and ends after it has enough time to complete.
#### The selected segment is too short
LTX Retake requires at least 2 seconds. Expand the handles so the segment is 2 seconds or longer.
#### The selected segment is too long
Keep the selection under 20 seconds. For more control, split a long change into shorter beats and process them separately.
#### Processing fails on the source clip
Try exporting the segment from Premiere as a short H.264 MP4, then use that as the Reshoot source. Chat Video Pro prepares incompatible formats when possible, but clean H.264 sources are the safest input.
***
### Links To Related Studio Pages
* [Studio](/features/studio) - Learn how Studio workflows are organized.
* [Erase Objects ](/features/studio/object-eraser-tool)- Remove unwanted objects from clips.
* [Rotoscope](/features/studio/sam-3-rotoscoping) - Keep a subject and remove the background.
* [Add Effects](/features/studio/kling-vfx) - Add VFX, weather, lighting, or style.
* [Relight Scene](/features/studio/relight-scene) - Change lighting or mood.
* [Upscale](/features/studio/video-upscaling) - Improve resolution after retakes or generation.
***
**Next:** If you need to remove a clearly selected object from a clip, use Erase Objects. If you need atmosphere or style, use Add Effects.
# Video Upscaling
Increase video resolution using AI upscaling models. Choose from Topaz, Flash VSR 2, or Bria models. Enhance quality for final delivery, upscale to 4K or 8K, and maintain quality during upscaling.
Upscale is the Studio workflow for increasing video resolution after a clip is generated, imported, cleaned up, or edited. Use it when the idea is working, but the final video needs more pixels, sharper detail, or a better delivery format.
Chat Video Pro supports multiple AI video upscalers, including **Topaz**, **Flash VSR 2**, and **Bria Video Increase Resolution**. Studio opens the same professional upscaling controls from a cleaner starting point, so you can load footage and go straight to resolution work.
***
### What This Tool Is For
Upscale is for **making a finished or near-finished video larger and cleaner**. It is usually the last step, not the first.
Use it for:
* Turning a generated 720p or 1080p clip into a higher-resolution delivery asset.
* Preparing AI results for client review, YouTube, social platforms, or a Premiere timeline.
* Improving apparent sharpness after Add Effects, Reshoot, Erase Objects, or Relight Scene.
* Matching lower-resolution inserts to a higher-resolution edit.
* Creating a cleaner master before compression.
* Testing whether older or lower-resolution footage can hold up in a modern project.
{% hint style="info" %}
Upscaling is not magic restoration. It can add detail, sharpen edges, and improve perceived resolution, but it works best when the source clip is already stable, clean, and worth finishing.
{% endhint %}
***
### When To Use It
Use Upscale when the clip content is already approved and the main issue is resolution or delivery quality.
Goal
Why Upscale helps
Finish an AI-generated clip
Raises the output resolution after generation.
Improve a post-production result
Makes cleanup or VFX results feel more finished.
Match a 4K timeline
Helps lower-resolution shots sit better beside 4K footage.
Prepare for client review
Gives the result a sharper, more polished presentation.
Create a higher-quality master
Lets you upscale before final export/compression.
Choose another Studio workflow first when:
Goal
Better workflow
Remove an unwanted object
Erase Objects
Keep a subject and remove the background
Rotoscope
Change a short section of the clip
Reshoot
Add atmosphere, VFX, or style
Add Effects
Change lighting or mood
Relight Scene
For most workflows, upscale after the creative edit is finished. If you upscale first, then edit, you may spend more time processing larger files and still need to upscale again at the end.
***
### Studio Path
The fastest route is through Studio.
1. Open **Studio**.
2. Choose **Upscale** from the Post-Production department.
3. Load a video from upload, Recents, or your Premiere timeline.
4. Choose the upscale model.
5. Choose a scale factor.
6. Pick codec or quality options when the selected model exposes them.
7. Review the output resolution preview.
8. Click **Generate Upscale**.
9. Compare before/after and click **Done** when you want to send it back to your chat/library.
Studio opens the video editor directly in Upscale mode. The default model is Topaz, and Studio disables models or scale factors when the requested output would exceed the model's limits.
***
### Classic/Editor Path
You can also upscale from the classic video editor path:
1. Import or generate a video.
2. Click **Edit** on the video thumbnail.
3. Choose **Upscale** from the model selector.
4. Configure model, scale, and output settings.
5. Generate and review.
The older **Transform** button path can also open upscaling controls from a video thumbnail. Use whichever path is closest to the asset you are already working with. For new finishing work, Studio is the cleanest starting point.
***
### Controls And Constraints
Control
What it does
Output Resolution
Shows the expected width and height before you generate.
Upscale Model
Chooses Topaz, Flash VSR 2, or Bria.
Scale Factor
Multiplies the source resolution by the selected factor.
Output Codec
Topaz exposes H.264 and H.265 options.
Output Quality
Flash VSR 2 exposes Low, Medium, High, and Maximum quality.
Generate Upscale
Starts the upscaling job.
Cancel
Stops an active upscale job when possible.
Current constraints:
Constraint
Detail
Source type
Video only.
Default model
Topaz.
Common scale factors
1.5x, 2x, 3x, and 4x, depending on model and source resolution.
Topaz output limit
Up to 4K output.
Flash VSR 2 output limit
Up to 4K output.
Bria output limit
Up to 8K output, with a 30-second video limit.
Reference images
Not used. Upscale works from the source video.
If a model or scale factor is disabled, it usually means the output would exceed the model's maximum resolution or duration limit.
***
### Choosing A Model
#### Topaz
Topaz is the best default for most users.
Use Topaz when:
* You want reliable, professional upscaling.
* You need a straightforward 1.5x, 2x, 3x, or 4x upscale.
* You are preparing AI-generated video for review or delivery.
* You want H.264 for compatibility or H.265 for smaller files.
Topaz is usually the safest first pass because it balances quality, predictability, and simple controls.
#### Flash VSR 2
Flash VSR 2 is useful when you want more control over the upscale quality.
Use Flash VSR 2 when:
* You want to compare quality levels.
* You are working with footage that needs a more controlled enhancement.
* You want a 4K-capped model with a quality setting.
* You are testing which upscaler treats your footage best.
In the current Studio/video editor flow, Flash VSR 2 is kept compatible with the CEP workflow by outputting an MP4-friendly result.
#### Bria
Bria is the option to consider when you need the highest output ceiling.
Use Bria when:
* You need up to 8K output.
* Your clip is short enough for Bria's 30-second limit.
* You want to test a quality-focused upscale against Topaz.
* Your source is already clean enough to justify a large output.
Bria is not the best starting point for long clips because of the duration limit. For longer videos, use Topaz or Flash VSR 2 when available.
***
### Codec And Quality Choices
#### H.264
Use H.264 when you want the safest, most compatible output.
Best for:
* Review links.
* Social delivery.
* General Premiere workflows.
* Smaller, easier-to-share files.
#### H.265
Use H.265 when you want better compression and your editing/delivery environment supports it.
Best for:
* Smaller files at higher resolution.
* Modern devices and platforms.
* Storage-conscious delivery.
#### Flash VSR 2 Quality
Flash VSR 2 exposes quality levels:
Quality
Best for
Low
Fast tests.
Medium
Balanced previews.
High
Default high-quality output.
Maximum
Slowest pass when quality matters most.
Use **High** first. Move to **Maximum** only when the source is worth the extra processing time.
***
### Best Practices
#### Upscale Last
Do creative work first, then upscale. This keeps iteration faster and avoids processing large intermediate files repeatedly.
Good order:
1. Generate, edit, clean up, or relight the clip.
2. Review the creative result.
3. Upscale the approved version.
4. Export or continue finishing in Premiere.
#### Start With 2x
2x is the most useful first test. It gives a clear quality bump without pushing the model as hard as 3x or 4x.
#### Watch For Sharpened Artifacts
Upscaling can make compression blocks, noise, flicker, motion blur, and AI artifacts more visible. Review faces, text, hands, edges, and high-motion areas after the upscale.
#### Match The Destination
Upscale for the project you are actually delivering:
* 1080p social ad: 1.5x or 2x may be enough.
* 4K timeline: 2x from 1080p is usually the practical target.
* Large display or special delivery: test Bria or a larger scale if the source supports it.
#### Do Not Use Upscale To Fix A Bad Generation
If the motion is wrong, the subject is inconsistent, or the VFX pass failed, upscale will only make the problem clearer. Fix the creative issue first, then upscale.
***
### Troubleshooting
#### A model is disabled
The source video may be too long or the requested output may exceed the model's limits. Bria is limited to 30 seconds, and Topaz/Flash VSR 2 are capped at 4K output.
#### A scale factor is disabled
The output resolution would be too large. Choose a lower scale factor or a different model.
#### The upscale is slow
Use a lower scale factor, a shorter clip, or a faster quality setting. Higher resolution outputs take longer to process.
#### The result looks sharper but worse
The source may contain compression, noise, flicker, or AI artifacts. Try a lower scale factor, use a cleaner source, or fix the creative issue before upscaling.
#### The file is too large
Use H.264 for compatibility or H.265 for smaller files when your workflow supports it. Avoid unnecessary 4x upscales when the delivery target does not need them.
#### The output does not match expectations
Try another model. Topaz, Flash VSR 2, and Bria can treat detail, edges, and motion differently. A one-model test is not always enough for important delivery work.
***
### Links To Related Studio Pages
* [Studio](/features/studio) - Learn how Studio workflows are organized.
* [Cinematic Lab](/features/studio/cinematic-lab) - Generate higher-quality source frames before video work.
* [Motion Director ](/features/studio/motion-director)- Animate still images before upscaling.
* [Add Effects ](/features/studio/kling-vfx)- Add VFX or style before the final upscale.
* [Reshoot ](/features/studio/reshoot)- Change a selected segment before upscaling.
* [Erase Objects](/features/studio/object-eraser-tool) - Clean a short clip before upscaling.
* [Relight Scene](/features/studio/relight-scene) - Change lighting before the final upscale.
***
**Next:** Upscale after the creative result is approved. If the shot still needs cleanup, use Erase Objects, Reshoot, or Add Effects first.
# Motion Capture
Transfer movements from any video to any character image. Create animations, place yourself in movie scenes, or bring illustrations to life—all by combining a motion video with a character image.
{% embed url="" %}
Motion Capture is the Studio workflow for taking movement from one video and applying it to a character image. The motion video provides the performance. The character image provides the identity. Chat Video Pro combines them with **Kling Motion Control** to create a new video where your character performs the reference motion.
***
### What This Tool Is For
Motion Capture is for **motion transfer**, not general video editing. You give Studio two assets:
Input
Purpose
Motion video
The source movement, gesture, pose timing, dance, walk, action, or performance.
Character image
The person, character, illustration, avatar, mascot, or design that should perform the motion.
The output is a new animated video. The character image is the visual anchor, while the motion video tells the model how that character should move.
{% hint style="info" %}
**The motion video is not the final look.** It is the choreography reference. The character image is what the viewer should recognize in the result.
{% endhint %}
Motion Capture is especially useful for:
* Animating a static character design.
* Making an avatar dance, walk, wave, teach, or perform.
* Turning a brand mascot into short-form content.
* Testing motion ideas before a proper shoot.
* Applying a real performance to an illustrated, cinematic, anime, or stylized character.
* Creating multiple clips with the same character and different motions.
***
### When To Use It
Use Motion Capture when the main creative question is:
> "Can this character perform that motion?"
Choose a different Studio workflow when the goal is not motion transfer:
Goal
Better workflow
Add rain, fire, atmosphere, or style to existing footage
Add Effects
Change lighting or mood on an image or video
Relight Scene
Generate a new moving shot from a still image
Motion Director
Create alternate camera angles
Multi-Cam
Remove a background or isolate a subject
Rotoscope
Remove unwanted people or objects
Erase Objects
Motion Capture works best when the motion is human-readable: dancing, walking, posing, gesturing, fighting, exercising, teaching, or acting. It is weaker when the source motion is hidden, heavily obstructed, very chaotic, or depends on objects the model cannot infer from the character image.
***
### Studio Path
The Studio path is the cleanest way to use Motion Capture because it presents the workflow as the dual-input process it actually is.
1. Open **Studio**.
2. Choose **Motion Capture** from the Post-Production department.
3. Add a **motion video** in the first slot.
4. Add a **character image** in the second slot.
5. Click **Continue**.
6. Studio uploads both assets and starts generation automatically.
7. Review the animated result.
8. Add it to your chat/library when it is ready.
The motion video can come from upload, Recents, or your Premiere timeline. The character image can come from upload, Recents, or frame capture.
Studio currently optimizes the key settings automatically:
* **Quality:** Pro.
* **Audio:** Original sound from the motion video is kept.
* **Prompt:** A focused baseline prompt is sent so the model follows the reference motion naturally.
* **Orientation:** Studio chooses the best mode based on the motion video duration.
{% hint style="success" %}
**This workflow is intentionally light on controls.** The quality of the two inputs matters more than tweaking settings. Spend your time choosing a clear motion video and a strong character image.
{% endhint %}
***
### Classic/Editor Path
The older model-selector path may still appear in parts of Chat Video Pro:
1. Import a motion video.
2. Select **Kling Motion Control** from the video model controls.
3. Attach one required character image.
4. Choose quality and character-orientation settings if they are exposed.
5. Generate the result.
That path is useful if you are already inside a classic chat/model flow. For new work, use Studio. The Studio version removes setup friction and makes the required two-input structure much harder to miss.
***
### Controls And Constraints
Control or constraint
Detail
Motion video
Required. Use 3-30 seconds. The performer should be visible and unobstructed.
Character image
Required. Use a clear image with visible face/body or a readable character silhouette.
Model
Kling Motion Control. Studio uses the Pro path for this workflow.
Character orientation
Auto-selected. Shorter videos are treated differently from longer motion-transfer clips.
Audio
The original sound from the motion video is kept.
Output aspect ratio
Follows the character image, not a hard-coded vertical frame.
File size
Motion video should stay within the workflow limit of about 200 MB.
#### How Auto Orientation Works
Kling Motion Control has two orientation behaviors:
Mode
What it is best for
Match Image
Keeps the character closer to the direction and framing of the character image. Best for shorter clips and image-driven animation.
Match Motion
Follows the motion video more aggressively. Best for longer dance, gesture, or body-performance references.
In Studio, this is chosen automatically:
* Videos around 10 seconds or shorter use the image-oriented path.
* Longer videos use the motion-oriented path, up to the 30-second Studio limit.
You do not need to choose this manually in the Studio workflow.
***
### Choosing Strong Inputs
#### Motion Video
The motion video should make the movement easy to understand.
Good motion videos usually have:
* One primary performer.
* Clear full-body or upper-body visibility.
* Good lighting.
* Limited occlusion.
* A simple enough background for the body motion to read.
* A clear beginning and ending pose.
* 3-10 seconds for testing, then longer clips once you trust the pairing.
Avoid motion videos where:
* The performer leaves the frame.
* Arms, legs, or the face are constantly blocked.
* The camera shakes heavily.
* Multiple people overlap.
* The movement depends on a prop that is not present in the character image.
* The action is too fast to read.
#### Character Image
The character image should tell the model who is performing the motion.
Good character images usually have:
* A clear face or head shape.
* Visible torso and limbs if the motion is body-heavy.
* A strong silhouette.
* Minimal occlusion.
* Enough resolution to preserve identity.
* A pose that is not wildly incompatible with the motion reference.
For dance, martial arts, workouts, or big gestures, a full-body or three-quarter character image is usually better than a tight headshot.
***
### Best Practices
#### Start With A Short Test
Do not start with a 30-second final clip. Test the pairing first.
1. Choose a 3-5 second section of motion.
2. Use the exact character image you plan to animate.
3. Generate once.
4. Check identity, pose stability, and motion transfer.
5. If the pair works, try a longer motion clip.
Short tests save time and make it easier to diagnose whether the motion video or the character image is causing issues.
#### Match Body Visibility To The Motion
If the motion is full-body dance, use a character image with the full body visible. If the motion is talking, teaching, waving, or acting from the waist up, an upper-body character image can work well.
The more the result needs legs, feet, hands, or torso rotation, the more those features should be visible in the character image.
#### Use Cinematic Lab For Better Character Images
Motion Capture is only as strong as the character image. For a polished result, create or refine the character first.
Useful prep workflows:
* Use Cinematic Lab to create a clean cinematic character still.
* Use Relight Scene to improve the lighting or mood of a character image.
* Use a saved character Element or consistent design image if you are building a series.
#### Do Not Fight The Motion
The baseline prompt tells the model to animate the character naturally using the reference motion. If you use the classic path where prompt text is exposed, keep the prompt aligned with the motion video.
Good:
{% code overflow="wrap" %}
```
A stylized mascot character performing the dance routine from the reference video in a clean studio setting.
```
{% endcode %}
Less effective:
```
A character walking slowly through a city street.
```
The second prompt contradicts a dance reference, so the model has competing instructions.
#### Keep A Character Sheet For Series Work
If you are making a run of clips with the same mascot, avatar, or fictional character, keep a small folder of approved character images. Use the same strongest image as the character input when possible. This makes the series feel more consistent than reinventing the character on every clip.
***
### Example Ideas
#### Brand Mascot Dance Clip
* Motion video: A clean 5-second dance reference.
* Character image: Mascot standing full body.
* Result: A short social animation where the mascot performs the dance.
#### Animated Teacher
* Motion video: A person gesturing while explaining something.
* Character image: Educational avatar or presenter character.
* Result: A teaching-style clip with natural hand and body movement.
#### Anime Performance
* Motion video: Performer doing a simple choreographed move.
* Character image: Anime-style character from Cinematic Lab.
* Result: Anime character performing realistic motion.
#### Fitness Demonstration
* Motion video: Trainer demonstrating a form or exercise.
* Character image: Branded coach or fitness avatar.
* Result: Character demonstrates the same motion for a tutorial or ad.
#### Character Reaction Pack
* Motion videos: Wave, point, shrug, celebrate, dance, and present.
* Character image: Same consistent character image.
* Result: A reusable library of animated character beats.
***
### Troubleshooting
#### The character does not look right
Use a clearer character image. The character should be large enough in frame, not blocked by objects, and visually readable at a glance. If the face or body design matters, make sure those features are visible.
#### The motion does not transfer well
Use a shorter, cleaner motion video. The model needs to understand the body movement. Avoid overlapping people, extreme blur, heavy camera shake, or clips where the performer leaves frame.
#### The body pose feels unstable
Try a character image with more of the body visible. A headshot gives the model less information about arms, torso, legs, clothing, and proportions.
#### The output framing looks wrong
Use a character image with the framing you want. Studio saves the result using the detected aspect ratio of the character image, so the character image strongly influences how the final video is displayed.
#### Audio is missing
Audio comes from the motion video. If the motion video has no useful audio, the generated clip will not magically create a soundtrack. Add music, effects, or voiceover in Premiere afterward.
#### Generation failed
Check that both inputs are present, your FAL API key is configured, the motion video is accessible, and the clip is within the Studio duration/file limits. If the problem persists, try a shorter motion clip and a smaller character image file.
***
### Links To Related Studio Pages
* [Studio](/features/studio) - Learn how Studio workflows are organized.
* [Cinematic Lab](/features/studio/cinematic-lab) - Create stronger character stills before animating them.
* [Relight Scene](/features/studio/relight-scene) - Improve lighting or mood on a character image.
* [Motion Director](/features/studio/motion-director) - Animate a still image with directorial camera motion instead of body motion transfer.
* [Add Effects](/features/studio/kling-vfx) - Add atmosphere, VFX, or character swaps to existing footage.
* [Multi-Cam](/features/studio/multi-cam) - Generate alternate views from images or videos.
***
**Next:** If you want to animate a still shot with camera movement instead of transferring body motion from a reference video, use Motion Director.
# Relight Scene
Relight Scene lets you change the lighting of an existing image or short video without rebuilding the shot from scratch. Use it to add golden hour, create a cinematic grade, move the key light, turn a flat frame into a moody scene, or test lighting directions before committing to a look in the edit.
{% hint style="info" %}
**Relight Scene is for changing light, not changing the shot.** The best results preserve the same subject, pose, camera angle, background, and composition. You are asking the model to relight the scene, not redesign it.
{% endhint %}
#### When to Use Relight Scene
Use Relight Scene when the shot is structurally right, but the lighting is not.
* **Make a flat frame feel cinematic** with contrast, mood, and direction
* **Turn midday into golden hour,** perfect for real estate, travel, or lifestyle edits
* **Create a clean studio look** from a usable but uneven product or portrait frame
* **Add a backlight or rim light** to separate a subject from the background
* **Preview lighting directions** before sending the image into Motion Director or AI Transitions
* **Match a generated still to the tone of the edit** before turning it into video
* **Apply a relit reference frame back onto a short video** so the moving clip follows the same lighting idea
Relight Scene is strongest when the original frame already has good composition and recognizable detail. If the frame is blurry, badly exposed, heavily compressed, or missing important subject detail, relighting cannot fully rescue it.
***
#### How the Workflow Works
Relight Scene has two paths:
Input
What Happens
Best For
Image
The image opens directly in the lighting configuration step, then generates a relit image
You pick one representative frame, relight that frame, then optionally apply the lighting to the whole short video
Short clips where one lighting look should carry through the shot
#### Getting Started
**Step 1: Add an Image or Video**
Relight Scene opens with an asset loader that accepts an image or a video.
You can start from:
* **Upload**
* **Recents**
* **Timeline / frame sources supported by the Studio loader**
Use an image when you want a single finished still. Use a video when you want to relight a moving shot.
**Step 2: If You Use Video, Pick the Hero Frame**
For video, Relight Scene first asks you to scrub through the clip and capture a frame.
This captured frame becomes the lighting reference. The workflow relights that frame first, then uses it as the visual guide when applying the look back to the video.
Pick a frame that shows:
* The main subject clearly
* The important background elements
* The lighting problem you want to solve
* A representative moment from the clip, not a motion-blurred in-between frame
{% hint style="info" %}
**Pro tip: choose the frame you would use as the thumbnail.** If the lighting looks good on that frame, the video pass has a better reference for the whole clip.
{% endhint %}
#### Direction vs. Style
Relight Scene has two modes:
Mode
Use It When
Result
Direction
You want to move the light source
Custom multi-light rig: Key, Fill, Hairlight, and Accent lights positioned freely around the subject
Style
You want a complete lighting mood
Golden hour, studio, overcast, daytime, nighttime, cinematic
Direction is about **where the light comes from**. Style is about **what the shot should feel like**.
***
#### Direction Mode
Direction mode gives you an interactive **3D light stage**. Your image appears centered in a stage with orbit rings — visual guides representing the sphere of directions a light can occupy. Drag a **light gizmo** anywhere on that sphere to set the light's direction: in front of the subject, to the side, above, below, or behind.
**Lighting Controls Panel**
Expand the **Lighting Controls** panel below the stage to configure the selected light:
Control
Options
What It Does
Type
Key, Fill, Hairlight, Accent
Sets the cinematic role of the light in the rig
Color
Color picker + H / S / L sliders
Neutral white for standard lighting; colored for gels or creative looks
Depth
In Front / Behind
Places the light on the near or far side of the subject on the Z axis
Intensity
0–100% slider
Controls brightness relative to the other lights in the rig
**Light Types**
Type
Role
When to Use
Key
The dominant main light — defines shadow direction and subject form
Start every rig with a Key; it establishes the direction and character of the whole look
Fill
A softer secondary light that lifts the shadows the Key creates
Add a Fill to prevent Key shadows from going too dark or harsh; keep intensity below the Key
Hairlight
A rim or back light above-behind the subject that separates it from the background
Use on any talking-head or portrait where the subject risks blending into the background
Accent
An edge or color light for creative highlights
Use with a colored gel for neon, editorial, or dramatic looks; or as a neutral edge for added depth
Tap **+ Add Light** to add a second or third light. Each light is configured independently. Up to **3 lights** are supported in a single rig.
**Two-Step Preview Workflow**
After configuring your rig, Direction mode generates a **lit still preview** first. Review the still, tweak light positions, types, colors, or intensity, and regenerate as many times as you need. Once the still looks right, click **Apply Lighting to Video** to apply the rig to the full clip.
**Rig Guidance**
**Build around the Key first**
Start with a single Key light. Position it where the main illumination should come from and get the angle and intensity right before adding more lights.
**Add a Fill to control shadow depth**
Place a Fill on the opposite side of the Key to lift the shadow side without flattening the look. Keep Fill intensity below the Key — typically 30–50%.
**Use a Hairlight for subject separation**
Position a Hairlight above-behind the subject. A neutral or slightly warm white at moderate intensity creates a thin highlight along the hair and shoulders that keeps the subject reading clearly against the background.
**Use Accent lights for color**
Set an Accent light to a deep blue, amber, magenta, or any creative hue using the Color picker. Keep intensity moderate so it complements — rather than overpowers — the Key and Fill.
***
#### Style Mode
Style mode applies a complete lighting atmosphere.
Style
Best For
Look
Golden Hour
Travel, lifestyle, real estate, beauty, outdoor scenes
Soft diffused light, low contrast, no harsh shadows
Daytime
Bright social edits, upbeat content, outdoor scenes
Clear natural daylight, vibrant color, high-key energy
Nighttime
Mystery, tension, cool moods, dramatic scenes
Blue moonlight, deep shadows, low-key darkness
Cinematic
Trailers, music videos, narrative edits, dramatic posts
High contrast, teal shadows, warm highlights, moody grade
#### Model Notes
Relight Scene uses two models depending on what you generate.
Stage
Model
Purpose
Relit Image
Nano Banana 2
Relights the captured image or source still
Relit Video
Kling O3 VFX (kling-o3-pro-edit)
Applies the approved lighting look to the original short video
**Image Relighting**
The relit image is usually the most controllable part of the workflow. You can regenerate the still until the lighting idea works before spending time on the video pass.
**Video Relighting**
For video, the workflow first creates a relit image. If you like it, click **Apply Lighting to Video**.
Kling O3 VFX receives:
* The original source video
* The relit image as a visual reference
* A lighting prompt based on the selected direction or style
The video pass is more expensive and slower than the image pass, so treat the image result like a proof before applying it to the video.
{% hint style="warning" %}
**Approve the relit frame before generating the video.** If the still result has the wrong mood, wrong shadows, or a changed identity, the video pass will inherit that problem.
{% endhint %}
***
#### Working With Images
For images, the flow is simple:
1. Load an image
2. Choose Direction or Style
3. Generate Relit Image
4. Compare before and after
5. Click **Done** to save the image to chat
Use image relighting before:
* Motion Director, when you want to animate a better-lit still
* AI Transitions, when your start and end frames need a consistent mood
* Cinematic Lab iterations, when you have the right scene but want a better lighting direction
* Thumbnail or hero-image creation, when the subject is good but the shot lacks polish
***
#### Working With Video
For the video, the flow is:
1. Load a 3-10 second video
2. Scrub to a representative frame
3. Capture the frame for relighting
4. Choose Direction or Style
5. Generate the relit image
6. If the image works, click **Apply Lighting to Video**
7. Compare original vs. relit video
8. Click **Done** to save the relit video to chat
Video relighting works best when:
* The camera movement is simple
* The subject stays visible
* The lighting change is plausible across the whole clip
* The clip is already edited down to the moment you want
* The captured reference frame represents the full shot
It is weaker when:
* The shot has fast cuts
* The subject leaves frame
* The lighting changes dramatically inside the original clip
* The video is too dark, noisy, or motion blurred
* Different parts of the clip need different lighting designs
***
#### Pro Tips
**Start with the least destructive lighting change**
If you only need polish, start with Studio or Overcast in Style mode, or a front-centered or mild side Key light in Direction mode. Save Nighttime and heavy Cinematic looks for shots that can handle more stylization.
**Use relighting before animation**
If you plan to animate a still in Motion Director, relight it first. A strong source still gives the video model a cleaner visual target.
**Use a hairlight or rear-positioned light to separate subjects**
A Hairlight or rear-positioned light is one of the most useful creative fixes. It can turn a flat subject/background relationship into something with depth without changing the composition.
**Use Overcast to normalize mismatched outdoor footage**
When two outdoor shots have harsh lighting differences, Overcast can sometimes make them easier to cut together because it reduces hard shadows and color extremes.
**Use Golden Hour for warmth, not accuracy**
Golden Hour is a mood tool. It can make lifestyle, travel, real estate, and creator footage feel more premium, but it may not match physically accurate sun direction in every frame.
**Use Nighttime only when the frame has enough detail**
Nighttime needs information to work with. If the original is already dark, pushing it darker can lose the subject.
**Do not judge video relighting from a bad reference frame**
If the captured frame is motion-blurred, blinking, obstructed, or poorly composed, go back and capture a cleaner frame before relighting.
**Regenerate the image before regenerating the video**
The image pass is the cheaper creative loop. Get that right first, then apply it to the clip.
Example Ideas
**Flat Interview to Cinematic Key Light**
Setting
Choice
Input
Talking-head still or short interview clip
Mode
Direction
Rig
Key light at 45° to one side + Fill on the opposite side at 30–50% intensity
Goal
Add shape to the face while keeping the subject readable
**Product Shot to Studio Polish**
Setting
Choice
Input
Product frame with uneven lighting
Mode
Style
Preset
Studio
Goal
Create clean commercial lighting before using the shot in an ad
**Travel Shot to Golden Hour**
Setting
Choice
Input
Outdoor lifestyle or landscape frame
Mode
Style
Preset
Golden Hour
Goal
Add warmth, premium mood, and softer sunset energy
**Subject Lost in Background**
Setting
Choice
Input
Portrait or character frame with weak separation
Mode
Direction
Rig
Hairlight above-behind the subject; optionally add a Key for front shape
Goal
Add rim light so the subject stands out from the background
**Day Scene to Moody Night**
Setting
Choice
Input
Clear daytime frame with enough subject detail
Mode
Style
Preset
Nighttime
Goal
Build a darker, cooler, more mysterious version
***
#### Best Practices
**Keep expectations focused**
Relight Scene is not a full VFX compositor. It can convincingly change mood and direction, but it is not meant to replace masks, rotoscoping, manual grade work, or scene reconstruction.
**Preserve the original composition**
The workflow is built around identity and composition preservation. Avoid asking it to change wardrobe, pose, location, camera angle, or props.
**Use one strong lighting idea**
"Golden hour with cinematic nighttime studio backlight and overcast softness" is too many directions at once. Pick one lighting concept per generation.
**Avoid text-heavy frames**
Like most image and video generation models, relighting can disturb fine text, labels, UI, and small logos. Use product frames with large, clear branding if text must remain visible.
**Trim video first**
For video, send only the section you need. A tight 4-6 second shot is easier to relight than a 10 second clip with changing action.
**Check identity carefully**
The workflow uses prompts to preserve faces, hair, skin tone, clothing, expression, and background, but you should still inspect portraits closely before using the result in client work.
***
#### Troubleshooting
**The Generate button is disabled**\
Check that your FAL API key is configured in Settings.
**My video was rejected**\
The clip must be between 3 and 10 seconds and under 200 MB. Trim or compress it before loading it again.
**The relit image changed the face**\
Try a less aggressive setup. In Direction mode, a front-centered or gentle side Key light is safer than extreme angles. In Style mode, Studio, Overcast, and Daytime are safer than Nighttime or heavy Cinematic looks.
**The lighting looks too extreme**\
Regenerate with a softer style or reduce light intensity in Direction mode. In Style mode, Overcast and Studio are usually safer than Nighttime or Cinematic.
**The video pass does not match the still perfectly**\
Video relighting has to preserve motion while applying the look. If the reference still is good but the video is inconsistent, try a shorter clip or capture a more representative frame.
**The subject becomes too dark**\
In Direction mode, move the Key light toward a front or front-side position. In Style mode, Studio, Daytime, or Overcast are safer choices. Nighttime and rear-positioned lights can reduce subject readability if the original frame is already low exposure.
**The background changed too much**\
Use a less stylized mode and start from a cleaner source frame. Relight Scene tries to preserve the background, but very aggressive lighting changes can still reinterpret details.
**The result is close but not final-grade quality**\
Use Relight Scene to create the base lighting direction, then finish the grade in Premiere with normal color tools.
***
**Next:** Use Motion Director to animate a relit still, or use AI Transitions to bridge two frames with a matching lighting mood.
# Reframe
Reframe changes a video clip's aspect ratio while AI fills the new edges to match the scene, making it easy to repurpose footage for any platform.
{% hint style="info" %}
**Reframe runs on Luma Ray 3.2** for sharper output at lower cost. Both quality tiers support clips up to **30 seconds** — Draft renders at 540p, Final at 720p. The original clip is kept intact — the AI generates new content along the expanded edges to fill the new aspect ratio. Use manual **reposition and zoom** controls in Studio or the chat quick-modal when you want to fine-tune subject placement.
{% endhint %}
**When to Use Reframe**
Reframe is the right tool any time your footage needs to live in a different aspect ratio than it was shot in — and you want the new frame to look complete, not cropped.
* **Repurpose landscape content for vertical social** — convert 16:9 footage to 9:16 for TikTok, Instagram Reels, and YouTube Shorts without losing the subject
* **Adapt footage to platform specs** — quickly reframe a single master clip for every delivery format your project requires
* **Explore creative framing without reshooting** — try a square, ultra-wide, or portrait ratio to see how the shot holds in a new aspect
* **Reframe generated Studio content on the spot** — use the quick action button on any Studio result card to immediately repurpose what you just created
***
**Starting a Reframe**
There are three entry points — all lead to the same workflow.
* **From Studio** — navigate to Studio → Post-Production → Reframe, then load a clip
* **From a chat clip** — click the Transform icon on any video clip thumbnail and choose **Reframe Video**
* **From a Studio result** — tap the quick Reframe action button directly on any Studio result preview card
***
**Getting Started**
**Step 1: Load Your Clip**
Pick the video clip you want to reframe. Three sources are available:
* **Upload** — select a file from your computer (MP4, MOV, or other common formats; up to 100 MB)
* **Recents** — pick from any clip recently used or generated in Chat Video Pro
* **Import from Premiere** — pull a clip directly from your active Premiere Pro timeline
The clip loader is skipped when entering from the chat Transform menu or a Studio result button — your clip is already pre-loaded.
***
**Step 2: Choose a Target Aspect Ratio**
Select the ratio you want the output to be. Seven options are available:
Ratio
Name
Common Use
16:9
Wide
YouTube, desktop, broadcast
9:16
Vertical
TikTok, Reels, YouTube Shorts
1:1
Square
Instagram feed, social ads
4:3
Classic
Standard TV, Zoom
3:4
Portrait
Upright portrait format
21:9
Cinema
Ultra-wide cinematic
9:21
Tall
Extra-tall vertical
Your current clip's aspect ratio is automatically excluded from the list — converting a clip to the ratio it is already in would have no effect.
***
**Step 3: Choose Quality**
* **Draft** — faster processing, supports clips up to **30 seconds**. Use Draft to test your ratio and fill hint before committing to a full render.
* **Final** — best quality output, supports clips up to **10 seconds**. Use Final for delivery.
{% hint style="info" %}
Start with Draft to quickly iterate on ratio choice and fill hint, then switch to Final for your delivery render.
{% endhint %}
***
**Step 4: Add an Optional Fill Hint**
The fill hint is a short text description that guides what the AI generates in the newly added areas. Leave it blank for fully automatic fill, or type a brief phrase when you want to steer the output.
**Examples:**
* `forest background, seamless`
* `brick wall, warm light`
* `dark studio, soft gradient`
* `extend concrete floor`
{% hint style="info" %}
Keep fill hints short and descriptive — a few words or one short phrase. Describe what the new area should look like, not what the subject should do. Subject positioning is handled automatically.
{% endhint %}
***
**Step 5: Generate and Use the Result**
Click **Generate**. A progress indicator shows while Reframe processes the clip. When the preview appears:
* Play it back to review the new framing and fill
* Click **Insert to Timeline** to place it on the Premiere Pro timeline
* Click **Save to Library** to store it for later use
If the result isn't quite right, adjust the fill hint or quality tier and regenerate.
***
**How Reframe Works**
**Subject positioning** — The AI detects the primary subject and centers it in the new frame by default. Use the **reposition and zoom** controls in Studio or the chat quick-modal to adjust placement when the auto-center is not quite right.
**AI fill** — Reframe generates new content along the expanded edges to match the look and content of the scene. The original clip remains intact; the AI adds to it rather than cropping.
**Input limits** — Clips must be at least 1 second and no larger than 100 MB. Duration limits depend on the quality tier: up to 30 seconds for Draft, and up to 10 seconds for Final.
***
**Pro Tips**
**Stable shots work best**
Locked-off shots, talking heads, product shots, and slow b-roll give the AI a consistent background to extend. Fast camera movement and quick pans make it much harder to generate a seamless fill — use stable, steady clips whenever possible.
**Draft first, Final for delivery**
Draft quality is fast enough to run several iterations quickly. Test different fill hints and confirm the framing looks right before running a Final render. The extra processing time for Final is worth spending only once you are satisfied with the Draft result.
**Use the fill hint for complex backgrounds**
When the new areas are large or the background is detailed, a short fill hint almost always produces better results than leaving the field blank. Describe the environment simply: "outdoor park, sunlight" or "dark office, city lights through window."
**Trim long clips before using Final quality**
Final quality supports clips up to 10 seconds. If your source clip is longer, trim it to the segment you need in Premiere Pro before importing — you get a better result and a faster render.
**Pair Reframe with other Studio tools**
Generate a clip in Motion Director or AI B-Roll, then immediately Reframe it for a different platform using the quick action button on the result card. The Studio result entry point is the fastest way to repurpose generated content the moment it lands.
***
**Troubleshooting**
**The fill edges look inconsistent or shift between frames**\
The source clip likely has camera movement. Reframe works best on locked-off shots. Try a more stable segment of the clip, or use a shorter section where the camera is still.
**The subject is off-center in the output**\
The source clip may have an extreme or off-center composition. Trim the source to a tighter crop in Premiere Pro that better centers the subject before sending it to Reframe.
**The fill area looks wrong or doesn't match the scene**\
Add or adjust the fill hint to describe the background more specifically. Even a brief description like "extend the sky, blue, bright" gives the AI useful context when the source edges are complex.
**The Generate button is disabled**\
Reframe requires an API key configured in Settings. Go to Settings → API Keys, confirm your key is entered, and check that it is funded.
**Draft quality looks noticeably rough**\
Draft is designed for quick iteration, not delivery. Once the ratio and fill hint look right in Draft, run a Final quality render for the version you plan to use.
**My clip is over 10 seconds and I want Final quality**\
Final quality supports clips up to 10 seconds. Trim the clip in Premiere Pro to a 10-second or shorter segment, then reimport and try again. Use Draft quality if you need to reframe the full longer clip.
**Generation failed or timed out**\
Try again — generation can occasionally fail under high load. If failures continue, check your API key status in Settings and confirm the key is funded.
***
**Next:** Learn about AI Color Match — instantly match the color grade of any clip to a reference frame.
# Story Scribe
Story Scribe is the transcription stage that feeds Story Cutter. It produces a transcript with word-level timestamps and speaker labels.
If you only ever transcribe English with Premiere's built-in panel, Story Scribe gives you three big upgrades: **more languages**, **better word-boundary precision** (so Story Cutter's cuts land cleaner), and a **private offline option** that never touches the cloud.
***
### What Story Scribe is designed to do
Story Scribe exists for one job: to produce the **most accurate possible transcript** of your dialogue so that downstream Story Cutter edits land on exact word boundaries instead of guessing in the middle of a sentence.
* **Word-level timestamps.** Every word in the transcript carries its own start/end time. When Story Cutter trims a soundbite, the cut snaps to the boundary between two words — not somewhere in the middle.
* **Multilingual.** 25+ languages out of the box (English, German, Spanish, French, Italian, Portuguese, Dutch, Polish, Swedish, Norwegian, Danish, Finnish, Greek, Czech, Ukrainian, Russian, Turkish, Arabic, Hebrew, Hindi, Vietnamese, Indonesian, Korean, Japanese, Mandarin, Cantonese, Thai). Premiere's native transcription panel is English-heavy — Story Scribe is built for the rest.
* **Speaker labels (diarization).** When the audio has multiple speakers, Story Scribe assigns each line to a "Speaker 1 / Speaker 2 / …" label that you can rename inline before handing off.
* **Three engines.** Pick the right tradeoff between speed, accuracy, cost, and privacy on a per-job basis — no global setting to change.
* **Story Cutter handoff.** One click ships the finished transcript into a new chat with Story Cutter, with the transcript already attached and the sequence linked.
{% hint style="info" %}
**Story Scribe is the front end for Story Cutter.** If you already have a transcript that Premiere produced (or you got it from somewhere else), Story Scribe can also **convert** it into the format Story Cutter needs — see Source 4: Upload existing transcript below.
{% endhint %}
***
### Getting Started
#### Step 1: Open Story Scribe
From the Studio launchpad, click the **Story Scribe** card (audio department, violet accent). You'll land on the **Setup** screen with four controls:
1. **Attach…** — pick a source (timeline, in/out range, audio file, or existing transcript)
2. **Language** — tell the engine what language to expect, or let it auto-detect
3. **Engine** — pick between FAST, STUDIO, or PRIVATE
4. **Transcribe / Convert & Import** — kick it off
The **Transcribe** button only enables when you've picked a source, AND your chosen engine supports the language you picked. (More on language compatibility in the Engines section.)
***
### Source options
You have four ways to feed audio into Story Scribe. Each one is tuned for a different real-world workflow.
#### Source 1: Active timeline
> **Best for:** The full pass — you want a transcript of everything in your current sequence.
* Make sure the sequence you want to transcribe is the **active** one in Premiere (i.e. the one with focus in the Program monitor).
* Click **Attach…** → **Active timeline**.
* Story Scribe will export the audio of the entire active sequence to a temp WAV, then send it to your chosen engine.
This is the default workflow. Use this for podcasts, interviews, vlogs, and any project where you want the cutter to have the whole conversation to choose from.
#### Source 2: In/Out range
> **Best for:** Long sequences where you only want to work with one section.
* In Premiere, scrub to the start of the section you want and press **`I`** to set the in-mark.
* Scrub to the end and press **`O`** to set the out-mark.
* Come back to the Story Scribe panel — within \~1 second, the **In/Out range** option will light up in the **Attach…** menu, showing the approximate duration (e.g. `≈4m 30s`).
* Click **Attach…** → **In/Out range**.
{% hint style="success" %}
Story Scribe **respects the in-mark offset**. If your in-mark is at `00:05:00`, the transcript's first word will be timestamped at `00:05:00` (not `00:00:00`) — so when Story Cutter places cuts, they land in the right spot on your full timeline, not a phantom 5-minute-earlier position.
{% endhint %}
If the **In/Out range** option is greyed out, it means Premiere doesn't currently report both marks set on the active sequence. Re-set them with `I` and `O` — The panel polls once a second and re-checks immediately when it regains focus.
#### Source 3: Upload audio file
> **Best for:** Raw camera audio, podcast stems, voice memos, anything that lives as a file on disk rather than on a Premiere timeline.
* Click **Attach…** → **Upload audio file**.
* Pick an **MP3**, **WAV**, **M4A**, or **MP4** from disk.
* Story Scribe sends the file directly to the engine — no AME export needed.
Useful when the speakers you care about live in raw camera files you haven't even imported into Premiere yet, or when you want a transcript of a podcast you're about to cut up.
#### Source 4: Upload existing transcript
> **Best for:** You already have a transcript and just want to use it in Story Cutter.
* Click **Attach…** → **Upload existing transcript**.
* Pick an **SRT**, **VTT**, or **JSON** file from disk.
* Story Scribe normalizes it into the canonical format Story Cutter expects, then drops you straight into the **Review** screen.
* The button label changes from **Transcribe** to **Convert & Import** to make this clear — no engine call is made, no cost is incurred, and the engine selector is ignored.
Supports:
* **SRT** — standard subtitle/caption format
* **VTT** — WebVTT format
* **JSON** — Premiere Pro's exported transcript JSON
{% hint style="info" %}
**Why does my Premiere transcript work but lose precision?** Premiere exports sentence-level timing only. Story Scribe will accept it and pass it through, but Story Cutter will warn that cuts will snap to sentence boundaries instead of word boundaries — they may feel slightly loose. For tight cuts, re-transcribe with Scribe v2 to get word timing back. (See Precision badge below.)
{% endhint %}
***
### Choosing an engine
Story Scribe ships with three engines. The cards are colour-coded by the trade-off they represent.
Engine
Best for
Speed
Cost (est.)
Privacy
Fal Whisper v3
Single speaker, clean audio
Very fast
~$0.10 / hr
Cloud (Fal.ai)
Fal Scribe v2 ⭐
Multiple speakers, accents, noisy audio
Fast
~$0.48 / hr
Cloud (Fal.ai)
Local Whisper (Windows only)
Sensitive content, offline work
Hardware-dependent
Free
100% on your machine
#### Fal Whisper v3 — FAST
The cheapest cloud option. Sends your audio to Fal.ai's hosted Whisper v3 endpoint. Excellent for clean single-speaker audio (one person at the mic, no overlapping voices, low background noise).
**Use it when:** you've got a podcast monologue, a single-presenter tutorial, a voiceover record, or any "one clean voice" job, and you want a fast, cheap turnaround.
**Don't use it when:** your audio has multiple speakers talking over each other, heavy accents, or background music — accuracy can dip noticeably compared to Scribe v2.
**Language note:** Fal Whisper v3 doesn't currently support Cantonese (`yue`). If you pick Cantonese in the language dropdown, this card is auto-locked, and the panel routes you to Scribe v2 or Local Whisper.
#### Fal Scribe v2 — STUDIO (recommended)
The accuracy default. This is the engine most people should pick for most jobs — it's what we recommend for serious production work.
**Use it when:** you have multiple speakers, accents, code-switching, music in the background, or you just want the most accurate result available. It's also the engine that the Local Whisper hardware-warning modal will offer to switch you to for long clips.
**Cost example:** a 60-minute interview runs about **$0.48** total. A 4-hour podcast is about **$1.92**.
{% hint style="success" %}
**Default to Scribe v2 unless you have a specific reason not to.** The accuracy gap over Fal Whisper v3 is real for multi-speaker and noisy audio, and the cost difference is rarely meaningful for occasional jobs.
{% endhint %}
#### Local Whisper — PRIVATE
Runs `whisper.cpp` on your own machine. Audio never leaves your computer — there's no cloud upload, no API key, no usage tracking. Zero cost per minute.
{% hint style="warning" %}
**Windows only for now.** Local Whisper is available on Windows 10/11 (x64). macOS support isn't shipped yet — if you're on a Mac, the card will be locked with a "Windows only" badge and the panel will steer you to one of the cloud engines.
{% endhint %}
{% hint style="warning" %}
**Speed depends entirely on your hardware.** Local Whisper runs on your CPU. A 60-minute clip can take anywhere from **5 to 30 minutes,** depending on your processor — a modern multi-core desktop will be near the fast end, an older laptop or a small Mac mini-class machine will be near the slow end. **Results may vary.** The panel won't freeze while it runs — you can keep editing in Premiere — but if you're on a deadline and you don't already know how fast your machine is, the cloud engines are a safer bet.
{% endhint %}
**First-run setup.** The first time you pick Local Whisper, Story Scribe shows an inline tile that downloads the `whisper.cpp` binary + the language model. The download is a one-time hit (about a few hundred MB, depending on the model size); after that, it runs offline forever. Downloads can be resumed if interrupted.
**Long-clip pre-flight warning.** If you pick Local Whisper for a clip 10 minutes or longer, Story Scribe shows a modal that estimates the time, shows you the equivalent cost on Scribe v2 (usually a few cents), and offers a one-click switch. You can dismiss the modal with **Don't show again** once you know your machine's profile.
***
### Picking a language
The language dropdown above the engine selector controls how the engine interprets the audio.
* **Auto-detect** — let the engine figure it out from the first few seconds. Good for unknown source material; not recommended if you already know the language because manual selection is always more reliable.
* **Manual selection** — pick from the 25+ language list. Always preferred when you know the language up front.
When you change the language, the engine cards reactively re-evaluate:
* If an engine doesn't support the language you picked, its card locks with a **🔒 Not available for \[Language]** badge.
* The most common case: switching to **Cantonese** locks **Fal Whisper v3** — use **Scribe v2** or **Local Whisper** instead.
***
### The Transcribing screen
Once you click **Transcribe**, you'll see the live transcription HUD with:
* **A halo waveform animation** breathing in violet
* **A progress percent** with a live sub-stage label ("Encoding in Media Encoder…", "Transcribing chunk 2 of 5…", etc.)
* **An elapsed counter** showing how long the run has been going
You can cancel at any time with the **Cancel** button. If anything fails, the panel shows the error and offers a retry.
{% hint style="info" %}
**Long clips are chunked automatically.** For audio over 20 minutes, cloud engines split the file into 20-minute chunks, transcribe each in parallel, and stitch the results back together with correct timestamps. You'll see the chunk progress in the live HUD ("chunk 3 of 7…").
{% endhint %}
***
### The Review screen
When transcription finishes, you land on the **Review** screen with the full transcript displayed line-by-line.
#### The precision badge
In the header, you'll see one of two pills:
* **Word-precise** (green) — every word has its own timestamp. Story Cutter has full precision to land cuts on exact word boundaries.
* **Sentence-precise** (amber) — only sentence-level timing is available. Cuts will snap to the start/end of each sentence — usually fine, but cuts may feel slightly loose if the AI wants to trim mid-sentence.
The label also tells you whether speakers were detected (`· Speakers detected`) or whether it was treated as a single speaker (`· Single speaker`).
#### Click-to-jump
Click any **word** or **segment** in the transcript and Premiere's playhead snaps to that exact timestamp and **starts playing**. The same behaviour Story Cutter uses for timestamp links — gives you instant audible confirmation that the click landed where you expected. A small **"Snapping…"** pill briefly appears while the jump is in flight.
#### Renaming speakers
If diarization gave you `Speaker 1`, `Speaker 2`, etc., click any speaker label inline to rename it. The new name propagates to every line that speaker said, and it persists when you save the JSON or hand off to Story Cutter.
#### The three actions
* **Save as JSON** — write the transcript to disk in the canonical Premiere transcript JSON format. Use this when you want a sidecar file you can re-use later.
* **Re-run** — go back to Setup with your source still attached. Useful if Scribe v2 gave you a sentence-precise result on a transcript upload and you want to re-transcribe from the original audio for word precision.
* **Use in Story Cutter** ✨ — the primary action. Saves the transcript, opens a new chat with the **Story Cutter Assistant** conversation starter, and attaches the transcript automatically. You're one message away from a rough cut.
{% hint style="warning" %}
**Sentence-precise transcripts trigger a confirmation modal.** When you click **Use in Story Cutter** on a sentence-precise transcript, Story Scribe asks you to confirm — and offers a one-click re-run with Scribe v2 to upgrade to word precision. You can dismiss this if you know what you're doing.
{% endhint %}
***
### Tips for getting the best results
#### Audio quality matters more than engine choice
The single biggest factor in transcript accuracy is the **audio that goes in**. A clean lapel mic on Fal Whisper v3 will beat a noisy room mic on Scribe v2 every time. Before reaching for a better engine, ask:
* Is the speaker close to the mic?
* Is the background noise low?
* Are there overlapping speakers? (Even diarization struggles with cross-talk.)
* Is there music underneath the dialogue?
If the answer to any of those is concerning, pick **Scribe v2** — it's more forgiving.
#### Set the language manually when you know it
Auto-detect is convenient but can occasionally pick the wrong language on the first few seconds (especially if the speaker starts with a non-native word, music plays first, or there's a long silence). When you know the language, **just pick it from the dropdown** — accuracy is always at least as good and usually better.
#### Use in/out marks to scope long sequences
If you only need to transcribe a 5-minute section of a 90-minute timeline, **set in/out marks** instead of transcribing the whole sequence. You'll get the result faster, pay less (cloud engines) or wait less (Local Whisper), and avoid clutter in the review screen.
#### For sensitive content, go local
Interview footage of a source who needs anonymity, medical content, legal depositions, anything under NDA — Local Whisper sends nothing to a cloud. The audio is read from disk, processed by `whisper.cpp` on your CPU, and the transcript is written back to disk. No network call is made.
#### Default to Scribe v2, drop down to Fal Whisper v3 only when it's truly a single clean speaker
Scribe v2's accuracy advantage is biggest on:
* Multiple speakers (interviews, panels, conversations)
* Heavy accents
* Background music or noise
* Code-switching between languages
If your material has none of those things — a solo voiceover record, a single-presenter tutorial recorded in a treated room — Fal Whisper v3 will save you about 5× on cost with negligible accuracy loss.
#### Don't re-transcribe what you can convert
If you've already got an `SRT`, `VTT`, or a Premiere transcript JSON sitting on disk, **use Upload existing transcript** instead of re-transcribing the audio. It's free, it's instant, and the only downside is sentence-level precision (which you can upgrade later with Re-run if Story Cutter needs the word boundaries).
#### Rename speakers before handoff
Story Cutter uses the speaker labels you see in the Review screen. Renaming `Speaker 1` → `Sarah` Before clicking **Use in Story Cutter** means Story Cutter's output will reference `Sarah` everywhere — much easier to scan than `Speaker 1`. Two seconds well spent.
#### Don't move clips on the source timeline after transcribing
Just like Story Cutter, Story Scribe locks timestamps to clip positions. If you transcribe a sequence, then re-arrange clips, then send the transcript to Story Cutter, the cuts will land on the **old** positions. Either transcribe last, or re-transcribe after any timeline reorganisation.
***
### Troubleshooting
**"In/Out range" stays greyed out.** Premiere doesn't currently report both an in-mark AND an out-mark on the active sequence. Re-press `I` and `O`, then click out and back into the Premiere panel — the Story Scribe poller will pick it up within 1 second.
**Local Whisper card is locked with "Windows only."** You're on macOS. Use Scribe v2 (recommended) or Fal Whisper v3 — both are cloud-based and run on any platform.
**The engine card is locked with "Language not supported."** The language you picked isn't in that engine's supported set. The most common case is Cantonese on Fal Whisper v3 — switch to Scribe v2.
**"Sentence-precise" badge after a re-run from audio** This usually means the audio path actually returned word timestamps, but the diarization step couldn't subdivide cleanly. Try re-running with Scribe v2 if you weren't already on it.
**The transcript looks correct, but the Story Cutter cuts feel loose.** Check the precision badge — if it's amber (sentence-precise), Story Cutter doesn't have word-level boundaries to snap to. Re-run with Scribe v2 from the audio source.
**Long clip on Local Whisper feels frozen** It's not — `whisper.cpp` is just slow on the CPU. The panel won't show per-second progress like the cloud engines because `whisper.cpp` it reports progress less granularly. Switch to Scribe v2 if you need faster turnaround.
***
### See also
* [Story Cutter Assistant](/conversation-starters/story-cutter-assistant) — the downstream tool that uses Story Scribe's transcripts to build rough cuts.
* [How to cut videos faster with AI-assisted story editing in Premiere Pro](/workflows/how-to-cut-videos-faster-with-ai-assisted-story-editing-in-premiere-pro) — end-to-end workflow combining Story Scribe + Story Cutter.
# Music Composer
{% hint style="info" %}
**Audio Expansion.** Music Composer is part of Chat Video Pro’s **Audio Expansion**. It is included with the **Creator Bundle** and **Creator Pass**, and available as a DLC add-on for **Base**. Generations bill to your own fal.ai account at wholesale rates — same as the rest of Studio.
{% endhint %}
Music Composer is the soundtrack studio inside Premiere. One Launchpad card, three modes: write an instrumental bed, build a full song with structure and vocals, or **Score a video** so the music follows the picture you attach.
**When to Use Music Composer**
* **Background beds** — underscore interviews, montages, or social cuts without hunting stock libraries
* **Original songs** — verse/chorus structure, voice casting, and lyrics when you need a vocal track, not just ambience
* **Picture-locked scoring** — attach a clip and generate a score timed to what is on screen
* **Stay in Premiere** — generate, audition, Add to Chat, and place on the timeline without bouncing through browser tabs
{% hint style="warning" %}
**Music Composer does not replace your DAW.** It creates finished audio (and for Score a video, an optional muxed preview) you can drop into the edit. Mix, duck, and refine in Premiere as usual.
{% endhint %}
**Getting Started**
**Step 1: Open Music Composer**
1. Open Chat Video Pro (**Window → Extensions → Chat Video Pro**)
2. Click **Studio** in the sidebar to open the Launchpad
3. In the **Audio** department, click the **Music Composer** card
You land on three mode tabs: **Instrumental**, **Song Builder**, and **Score a video**.
***
#### Mode: Instrumental
Best for beds, underscore, and lyric-free soundtrack.
**Step 2: Details — Describe your music**
* Write what you want to hear (mood, instruments, energy, use case). Aim for at least a clear sentence — short, vague prompts produce generic beds.
* Optional: enable the **background music** style when you want lower-energy underscore that sits under dialogue.
* Tap the sparkle (**AI Optimize**) to tighten the prompt, then **Continue**.
**Step 3: Style — Model & duration**
Pick an engine and a duration, then click **Generate Music**.
Model
Best for
Duration
Lyria 3 Pro
High-quality instrumental beds
Fixed length (duration control may not apply)
MiniMax Music 2.6
Flexible instrumental generation
Fixed length (duration control may not apply)
ElevenLabs Music
Prompt-driven soundtrack
About 5–150 seconds
Stable Audio 3
Longer beds and atmospheres
About 5–380 seconds
{% hint style="info" %}
If exact length matters for your cut, prefer a model that honors duration (ElevenLabs Music or Stable Audio 3), then trim in Premiere if needed.
{% endhint %}
**Step 4: Result**
Preview under **Your music**, then:
* **Add to Chat** — primary handoff into the conversation / Library
* **Place on timeline** — inserts at the **playhead** on a free audio track
* **Try Again** — regenerate with the same setup
***
#### Mode: Song Builder
Best when you want structure and vocals — not just a loop.
**Step 2: Details — Build your song**
* Add song sections (Intro, Verse, Pre-Chorus, Chorus, Post-Chorus, Hook, Bridge, Break, Outro).
* Fill lyrics per block (or use **Magic** to draft lyrics from your idea).
* Tap **AI Optimize** when you want help polishing lyric/style language, then **Continue**.
{% hint style="info" %}
MiniMax lyric path has a practical ceiling around **3,500 characters** of lyrics. If Generate stays disabled, shorten the lyric blocks or switch to ElevenLabs Music for the song path.
{% endhint %}
**Step 3: Style — Voice & model**
* Choose a vocal voice casting option.
* Pick **MiniMax Music 2.6** or **ElevenLabs Music**.
* Click **Generate Song**.
**Step 4: Result**
Same handoff as Instrumental: preview, **Add to Chat**, **Place on timeline** (playhead), **Try Again**.
***
#### Mode: Score a video
Best when the music should follow a specific clip — trailers, social cuts, picture-locked sequences.
**Step 2: Attach a clip**
The empty state opens the clip loader titled **Score a video**. Choose:
* **Upload** — video file from disk
* **Recent** — a video you used or generated recently in Chat Video Pro
* **Import Clip** — pull the selected clip from the active Premiere timeline (best when you want the score placed back under that clip)
{% hint style="warning" %}
**Max clip length: 180 seconds.** Longer clips are rejected until you trim or re-export a shorter section.
{% endhint %}
**Step 3: Optional style notes → Generate Score**
* Add short **Style notes (optional)** — genre, mood, instruments, or “sparse under dialogue.”
* Use **AI Optimize** if you want the notes tightened for scoring.
* Click **Generate Score** (shows **Preparing…** while the clip uploads and the score is planned).
Music Composer watches the clip, builds a composition plan behind the scenes, and generates a synced score. You do not edit the plan in this release — steer with style notes and regenerate if needed.
**Step 4: Result — Your scored video**
* Preview the muxed video with the score
* **Download video with score** — export the muxed preview
* **Download audio** — score-only file from the audio pill
* **Add to Chat** — promotes the audio and the muxed video into chat
* **Place on timeline** — **under the source clip** when you used **Import Clip** (timeline start was captured); otherwise at the **playhead**
* **Try Again**
**Tips for Best Results**
* Be specific about **role** (“sparse piano under interview,” “driving drums for montage,” “warm acoustic for brand film”).
* For Score a video, prefer **Import Clip** when the music must land under a known timeline position.
* Keep Score clips under three minutes; cut selects before attaching.
* Use **AI Optimize** when your prompt is a vibe dump — it turns loose notes into a usable brief.
* Audition in the panel, then place — Place does not auto-duck dialogue.
**Troubleshooting**
Symptom
What to try
Generate stays disabled
Prompt too short, lyrics over the MiniMax cap, or clip over 180s for Score
Duration doesn’t match what you set
Switch to ElevenLabs Music or Stable Audio 3; some engines ignore duration
Score feels off-picture
Tighten style notes, regenerate, or attach a cleaner / shorter select
Place lands at the playhead, not under the clip
Re-attach with Import Clip so timeline start is captured
Engine / network errors
Confirm fal.ai key + balance in Settings → Usage, then Try Again
**See also**
* SFX Studio — text SFX and Foley from video
* Stem Separation — split a finished track into Vocals, Drums, Bass, and more
* Voiceover Lab — narration and voice cloning
* Pricing FAQ — Audio Expansion entitlement
# SFX Studio
{% hint style="info" %}
**Audio Expansion.** SFX Studio is part of Chat Video Pro’s **Audio Expansion**. It is included with the **Creator Bundle** and **Creator Pass**, and available as a DLC add-on for **Base**. Generations bill to your own fal.ai account at wholesale rates — same as the rest of Studio.
{% endhint %}
SFX Studio is the sound-design card in the Audio department. Two modes: **Text** (describe a sound and generate it) and **Foley** (attach a video clip and generate synced sound design for what is on screen).
**When to Use SFX Studio**
* **Custom hits and atmospheres** — whooshes, impacts, UI clicks, room tone, creature vocals — without digging through SFX libraries
* **Picture-locked Foley** — generate sound that follows a clip you already cut
* **Quick variants** — try several takes of the same prompt before you commit to the timeline
* **Stay in Premiere** — generate, Add to Chat, and Place on timeline from one panel
{% hint style="warning" %}
**SFX Studio ships Text and Foley in this release.** Performance / live-transform modes and automated SFX spotting are not available yet — do not expect those tabs.
{% endhint %}
**Getting Started**
**Step 1: Open SFX Studio**
1. Open Chat Video Pro (**Window → Extensions → Chat Video Pro**)
2. Click **Studio** → **Audio** department
3. Click the **SFX Studio** card
Toggle **Text** or **Foley** at the top of the workflow.
***
#### Mode: Text
Best when you can describe the sound you need.
**Step 2: Prompt — Describe your sound**
* Write a concrete sound description (what happens, material, space, intensity).
* Prompt length is capped around **300 characters** — keep it specific, not essay-length.
* Optional: use **AI Optimize** to sharpen the wording, then **Continue**.
**Example prompts:**
> “Heavy wooden door slam in a stone hallway, short reverb tail”
> “Soft cloth rustle as someone sits on a leather couch, close mic”
> “Futuristic UI confirm beep, clean and short”
**Step 3: Model & settings → Generate SFX**
Pick a model and duration, then click **Generate SFX**.
Model
Best for
Duration guide
ElevenLabs SFX v2
General SFX, default pick
Up to about 22 seconds
CassetteAI SFX
Fast / alternate character
Up to about 30 seconds
Stable Audio 3
Longer atmospheres and beds
Up to about 120 seconds
You can request multiple variants in one run when the UI offers a variants control — audition each pill before placing.
**Step 4: Result**
* Preview each take
* **Add to Chat**
* **Place on timeline** — inserts at the **playhead** on a free audio track
* **Try Again**
***
#### Mode: Foley
Best when the sound must follow a video clip.
**Step 2: Attach a clip — Foley from video**
The empty state opens the loader titled **Foley from video**:
* **Upload** — video from disk
* **Recent** — a recent Chat Video Pro video
* **Import Clip** — selected clip from the active Premiere timeline (preferred when you want sync placement back under that clip)
{% hint style="warning" %}
**Max clip length: 60 seconds.** Trim longer selects before attaching.
{% endhint %}
After attach, you’ll see a clip preview. Optionally add a short prompt to steer materials or intensity (some Foley engines require a non-empty prompt — leave a simple ambient note if you’re unsure).
**Step 3: Model → Generate Foley**
Model
Best for
HunyuanVideo-Foley
Higher-quality picture-aware Foley (default)
ThinkSound
Faster Foley passes
Click **Generate Foley** (shows **Preparing…** while the clip is prepared and uploaded).
**Step 4: Result**
* Preview the Foley audio
* **Add to Chat**
* **Place on timeline** — **under the source clip** when you used **Import Clip**; otherwise at the **playhead**
* When Import Clip was used, you may also see **Place at playhead** if you want the override
* **Try Again**
**Tips for Best Results**
* Describe **physical events**, not abstract moods (“glass shatter on tile” beats “dramatic sound”).
* For Foley, use **Import Clip** when placement under the source shot matters.
* Keep Foley selects under a minute; shorter clips score more reliably.
* Generate two or three Text variants before you commit — first takes are often “close but not cut.”
* Place on a free track, then duck or EQ against dialogue in Premiere.
**Troubleshooting**
Symptom
What to try
Generate disabled / preparing forever
Clip over 60s (Foley), empty prompt when required, or fal.ai key/balance issue
Foley feels unsynced
Re-attach a cleaner select; try the other Foley model; Place under Import Clip source
Place lands at playhead unexpectedly
Use Import Clip so timeline start is captured, or use Place at playhead intentionally
Text result too long / too short
Pick a model whose duration range matches the hit you need
**See also**
* Music Composer — instrumentals, songs, and Score a video
* Stem Separation — isolate vocals and instruments from a finished track
* Voiceover Lab — dialogue and narration
* Pricing FAQ — Audio Expansion entitlement
# Voiceover Lab
{% hint style="info" %}
**Audio Expansion.** Voiceover Lab is part of Chat Video Pro’s **Audio Expansion**. It is included with the **Creator Bundle** and **Creator Pass**, and available as a DLC add-on for **Base**. Generations bill to your own fal.ai account at wholesale rates — same as the rest of Studio.
{% endhint %}
Voiceover Lab turns a script into spoken narration without leaving Premiere. Write (or paste) copy, pick a catalog voice or one of **My Voices**, generate, then **Add to Chat** and drop the take onto your timeline.
**When to Use Voiceover Lab**
* **Narration and explainers** — product demos, tutorials, ads, social VO
* **Temp VO for picture lock** — placeholder reads before a talent session
* **Consistent brand voice** — reuse a cloned voice across projects
* **Avatar pairing** — the same voice layer powers Avatar Studio when you need a talking head on camera
{% hint style="warning" %}
**Voiceover Lab is chat-first.** Results go to **Add to Chat** — there is no Place on timeline button in this workflow. Drag the audio from chat into Premiere (or place from Library) when you’re ready.
{% endhint %}
**Getting Started**
**Step 1: Open Voiceover Lab**
1. Open Chat Video Pro (**Window → Extensions → Chat Video Pro**)
2. Click **Studio** → **Audio** department
3. Click the **Voiceover Lab** card
You’ll see three steps: **Script** · **Voice** · **Generate**.
***
#### Step: Script — Write your script
* Type or paste your narration into the script editor.
* The Script step allows up to about **10,000 characters** for drafting.
* Tap the sparkle (**Optimize script…**) to tighten pacing, clarity, or delivery notes, then review and keep or dismiss the suggestion.
* Click **Continue**.
{% hint style="info" %}
Write the way you want it spoken. Short sentences and explicit pauses (“…”) read more naturally than dense paragraphs.
{% endhint %}
***
#### Step: Voice — Choose a voice
* Browse the voice catalog (filter by gender, accent, tone, use-case, or provider).
* Preview a voice before you commit.
* For MiniMax voices, choose **HD** or **Turbo** quality when offered.
* Open **My Voices** to use a cloned voice, or create one from a recording / upload (aim for a clean sample of about **10+ seconds**).
**Character limits at generate time** (enforced for the selected voice’s provider):
Provider
Max characters
ElevenLabs
5,000
MiniMax
10,000
If Generate is blocked, shorten the script or switch to a provider with a higher cap.
{% hint style="info" %}
**Cloned voices can expire** on the provider side after roughly a week of inactivity. If a custom voice fails with a “not found” style error, Voiceover Lab can refresh/reclone automatically when you regenerate — keep your original sample handy.
{% endhint %}
Click **Generate Voiceover** when the script and voice are set.
***
#### Voice cloning (My Voices)
Clone a custom voice from a short sample, then use it like any catalog voice in Voiceover Lab (and in Avatar Studio — they share the same voice layer).
{% hint style="info" %}
**Requirements:** a clean audio sample of at least **10 seconds**, and a connected **fal.ai** key (Settings). Cloning and speech generations bill to your fal.ai account at wholesale rates.
{% endhint %}
**When to clone a voice**
* You want narration that sounds like **you** (or a client / talent) instead of a preset
* You need the **same voice** across Voiceover Lab takes and Avatar Studio talking heads
* You’re building a reusable brand VO without booking a booth every time
**How to clone a voice**
**Step 1: Open the clone flow**
1. Open **Voiceover Lab** → go to the **Voice** step
2. Click **Clone Voice** (primary green button), **or** open **Manage Voices** and switch to the add/clone flow
The **Manage Voices** modal title explains the rules: clone from a **10s+** sample; cloned voices **expire after 7 days** and **auto-reclone on use**.
**Step 2: Name the voice**
* Give it a clear name (e.g. “Client Brand VO” or “Client – Sarah”)
* Optional tags help you find it later under **My Voices**
**Step 3: Provide a sample (pick one source)**
Source
How
Record
Use the in-panel recorder. A guided transcript appears so you have something natural to read — click Start recording, speak clearly, stop when you have 10+ seconds.
Upload
Choose an audio file (common types: MP3, WAV, M4A, OGG).
{% hint style="warning" %}
**Sample must be at least 10 seconds.** Shorter clips are rejected with a message showing your duration. Prefer a quiet room, dry mic, no music bed, and no heavy reverb.
{% endhint %}
**Step 4: Clone**
1. Confirm the sample looks ready (duration ≥ 10s)
2. Click **Clone Voice** (shows **Cloning…** while it runs)
3. When it finishes, the voice appears under **My Voices**
**Step 5: Use it**
1. On the Voice step, filter to **My Voices** (or find the card in the grid)
2. Select the cloned voice (preview if available)
3. Continue with your script and click **Generate Voiceover**
**Managing cloned voices**
In **Manage Voices** you can:
* Browse **My Voices (N)**
* Audition a sample (only one plays at a time)
* Delete a voice you no longer need (confirm in the panel — no system dialog)
**Expiry and auto-refresh**
Cloned voices are stored with the AI provider and typically **expire after about 7 days**.
* Chat Video Pro keeps your **local sample** so it can **auto-reclone** when you generate and the old voice id is gone
* If a custom voice fails with a “voice not found” style error, try **Generate** again — the panel should refresh the clone and retry
* Keep a backup of the original sample outside the panel if the voice matters for long-running projects
{% hint style="success" %}
**Best practice:** re-clone (or generate once) with an important voice every few days during active work so the provider copy stays fresh — or just rely on auto-reclone and keep the sample file.
{% endhint %}
**Tips for a good clone**
* Read in a natural speaking voice — not whispered, not shouted
* One speaker only; avoid background TV, fans, or music
* Phone voice memos are fine if the room is quiet and the level isn’t clipping
* Longer than 10s is OK; clarity matters more than length past the minimum
* Name voices by person + use case so you don’t mix “temp booth” with “final brand”
**Cloning troubleshooting**
Symptom
What to try
“Sample must be at least 10 seconds”
Record/upload a longer clip
Clone fails / can’t read media
Re-upload the file, or re-record in the panel (avoid broken paths)
Clone button stays disabled
Add a name, attach a valid sample ≥ 10s, confirm fal.ai key
My Voices is empty / filter hidden
You haven’t cloned yet — use Clone Voice
Custom voice fails on Generate
Usually expiry — generate again to auto-reclone; keep the original sample
Clone sounds wrong / noisy
Re-record dry and quiet; don’t use music beds or heavy FX
**Using a clone in Avatar Studio**
Cloned voices live in the shared voice library. After you clone in Voiceover Lab (or Manage Voices), open **Avatar Studio**, pick **Script + voice**, and select the same voice from **My Voices** so the talking head matches your VO.
***
#### Step: Generate — Result
* Preview the take in the result player
* **Try Again** — same voice, new take
* **Try Different Voice** — jump back to Voice without losing the script
* **Add to Chat** — hand off into the conversation / Library for timeline use
There is **no Place on timeline** control in Voiceover Lab by design.
**Tips for Best Results**
* Optimize after you have a solid draft — not before you’ve decided what the VO must say.
* Match provider to length: long scripts often fit MiniMax’s higher cap; short ads fit either.
* Clone from quiet, dry recordings — music beds and heavy reverb confuse the clone.
* For Avatar Studio, generate or pick the voice here first, then reuse it in the talking-head workflow.
* After Add to Chat, place VO on a dedicated track and nudge to picture in Premiere.
**Troubleshooting**
Symptom
What to try
Generate disabled
Script over the provider cap (5k / 10k), empty script, or no voice selected
Clone rejected
Sample too short (< ~10s), too noisy, or wrong file type — record/upload again
Custom voice suddenly fails
Provider expiry — regenerate to refresh; keep the original sample
No Place on timeline button
Expected — use Add to Chat, then drag into Premiere
Engine errors
Check fal.ai key + balance in Settings → Usage
**See also**
* Avatar Studio — talking-head video with the same voice layer
* Story Scribe — transcripts that can feed narration workflows
* Music Composer — underscore under your VO
* Pricing FAQ — Audio Expansion entitlement
# Stem Separation
{% hint style="info" %}
**Audio Expansion.** Stem Separation is part of Chat Video Pro’s **Audio Expansion**. It is included with the **Creator Bundle** and **Creator Pass**, and available as a DLC add-on for **Base**. Generations bill to your own fal.ai account at wholesale rates — same as the rest of Studio.
{% endhint %}
Stem Separation takes a finished audio file and splits it into isolated parts — vocals, drums, bass, and more — so you can remix, mute, or rebuild a mix without leaving Premiere.
**When to Use Stem Separation**
* **Pull vocals** — acapellas for edits, memes, or remixes
* **Instrumental beds** — remove vocals from a music track you already own rights to use
* **Remix in the timeline** — place Vocals / Drums / Bass / Other on separate tracks
* **Cleanup passes** — isolate a problem stem before repair or replacement
{% hint style="warning" %}
**Source audio only from Upload or Recent.** Stem Separation does not pull a clip directly from the Premiere timeline yet — export or save the audio first, then attach it here.
{% endhint %}
**Getting Started**
**Step 1: Open Stem Separation**
1. Open Chat Video Pro (**Window → Extensions → Chat Video Pro**)
2. Click **Studio** → **Audio** department
3. Click the **Stem Separation** card
You’ll see three steps: **Input** · **Stems** · **Generate**.
***
#### Step: Input — Source audio
Attach the track you want to split:
* **Upload** — audio file from disk
* **Recent** — audio you used or generated recently in Chat Video Pro
After attach, confirm **Source audio**, use **Change** if you picked the wrong file, then **Continue**.
***
#### Step: Stems — Stem options
Choose how many parts to extract, then click **Separate Stems**.
Option
You get
6 stems (default)
Vocals · Drums · Bass · Other · Guitar · Piano
4 stems
Vocals · Drums · Bass · Other
{% hint style="info" %}
Start with **6 stems** when guitar/piano isolation matters. Use **4 stems** for a simpler vocal/instrumental split.
{% endhint %}
Result pills label the engine as **Demucs (6-stem)** or **Demucs (4-stem)**.
***
#### Step: Generate — Results
Each stem appears as its own audio pill:
* Preview any stem
* **Place on timeline** — per stem, at the **playhead** on a free audio track
* **Add to Chat** — sends the full stem set into the conversation / Library
* **Try Again** — re-run with the same or adjusted stem count
**Where stems land in Library**
With the Audio Library update enabled, separated stems show under **Library → Audio → Stems**. They are **not** listed under Music — Music is for non-stem music items only.
**Tips for Best Results**
* Feed a clean stereo mix when you can — heavy limiting and stacked FX reduce isolation quality.
* Separate once, then place only the stems you need (e.g. Vocals + Other for a dialogue bed).
* Name or tag takes in chat so you can find the set later in Library → Stems.
* Rights still matter: separation does not grant rights to commercial music you don’t own or license.
**Troubleshooting**
Symptom
What to try
Can’t attach from the timeline
Expected — export/save audio, then Upload or pick Recent
Missing guitar/piano
Switch to 6 stems and Separate again
Can’t find stems in Library Music
Open Library → Audio → Stems
Bleed between stems
Normal for dense mixes — tighten with EQ/automation in Premiere
Engine errors
Check fal.ai key + balance in Settings → Usage, then Try Again
**See also**
* Music Composer — generate new music instead of splitting an existing track
* SFX Studio — create new sound effects and Foley
* Voiceover Lab — narration instead of pulled vocals
* Pricing FAQ — Audio Expansion entitlement
# Custom Templates
Custom Presets let you save your own Motion Director and AI Transitions recipes as reusable cards. Instead of rewriting the same camera move or transition prompt every time, you can create a named preset, add an optional thumbnail, set a recommended duration, and reuse it from the workflow picker.
{% hint style="info" %}
**Custom presets are for repeatable creative direction.** Use them when you have a camera move, transition style, brand look, or editorial trick you expect to use more than once.
{% endhint %}
#### Where Custom Presets Work
Custom presets are currently available in:
Workflow
What You Can Save
Extra Field
Motion Director
A reusable image-to-video camera prompt
None
AI Transitions
A reusable two-frame transition prompt
Optional negative prompt
Custom presets are workflow-specific and only appear in the workflow they were made in.
***
#### When to Use Custom Presets
Use custom presets when the built-in cards are close, but not specific enough.
* **Save a branded motion style** you use across a channel or client
* **Create specialty camera moves** that are not built into Motion Director
* **Build custom transition recipes** for a series, template pack, social format, or edit style
* **Keep duration choices consistent** for repeatable deliverables
* **Add transition negative prompts** when a custom transition keeps producing the same artifact
* **Turn successful experiments into reusable cards** after you find a prompt that works
Custom presets are not meant to replace the built-ins for normal work. Built-ins are still better when you want the workflow to analyze the image or frames and fill in the prompt automatically.
***
#### The Most Important Difference
Built-in presets and custom presets behave differently.
Preset Type
What Happens
Built-in Motion Director presets
The workflow analyzes the image, detects the subject, builds a movement-specific prompt, adds identity and stability language, and uses preset-specific generation settings
Custom Motion Director presets
Your saved prompt is sent directly as the motion prompt
Built-in AI Transition styles
The workflow analyzes the start and end frames, fills a transition template with subject/start/end/environment details, and adds quality-lock instructions
Custom AI Transition presets
Your saved prompt is sent directly as the transition prompt, with optional negative prompt if you saved one
#### How to Create a Custom Preset
1. Open **Motion Director** or **AI Transitions**
2. Click the movement or transition card to open the preset picker
3. Click **Custom**
4. Add a **Name**
5. Optional: upload a **Thumbnail**
6. Write the **Prompt**
7. For AI Transitions only: optional **Negative prompt**
8. Set the **Recommended duration**
9. Click **Save**
After saving, the custom preset appears in that workflow's picker alongside the built-in cards.
#### Editing and Deleting
Custom preset cards include an edit icon.
Use it to:
* Rename the preset
* Replace the thumbnail
* Update the prompt
* Change the recommended duration
* Add or revise a transition negative prompt
* Delete the preset
Deleting a preset is permanent.
***
#### Template Fields
Field
What It Does
Notes
Name
The card title in the preset picker
Keep it short and recognizable
Thumbnail
Optional image preview for the card
Useful for branded looks or visual transition types
Prompt
The exact instruction sent during generation
Required
Negative prompt
What AI Transitions should avoid
AI Transitions custom presets only
Recommended duration
The duration selected when the preset is chosen
3-15 seconds
The name is for you. The prompt is for the model. The duration is for the workflow.
***
#### Recommended Duration Guide
Use duration as part of the creative design.
Duration
Best For
3-4s
Whip moves, punchy transition hits, fast camera accents
5-6s
Product reveals, subtle motion, most social edits, clean transitions
8-10s
Morphs, time passage, orbit-style moves, complex reveals
10-15s
Slow atmospheric shots, long camera travel, complex scene transformations
If a prompt requires a lot of physical change, give it more time. If the move is just a hit, wipe, or quick camera accent, keep it short.
***
### Motion Director Custom Presets
Motion Director custom presets are for image-to-video camera direction.
Use them when you want to save a camera language that the built-in movement list does not cover exactly.
#### How to Write Motion Director Prompts
A good Motion Director custom prompt should include:
* **Camera behavior**: what the camera does over time
* **Subject lock**: what must stay unchanged
* **Scene dynamics**: optional motion inside the frame
* **Stability instructions**: what should not morph, drift, or appear
* **Shot tone**: documentary, commercial, cinematic, atmospheric, handheld, etc.
Because the custom prompt is used directly, write the prompt as a complete instruction.
Good structure:
{% code overflow="wrap" %}
```
[Camera move]. The subject in the source image remains the same person/object with the same clothing, features, and pose. Preserve the original background and lighting. Add [subtle scene motion]. No new characters, no morphing, no warping, no scene replacement.
```
{% endcode %}
Avoid:
* Placeholder-only prompts like `[Subject] moves forward`
* Asking for a new outfit, new location, or new character
* Combining too many camera moves at once
* Writing a still-image prompt instead of a video/camera prompt
{% hint style="info" %}
**Pro tip: custom Motion Director prompts should name the camera, not just the vibe.** "Slow parallax drift with foreground movement" is better than "cinematic and premium."
{% endhint %}
***
### Motion Director Example Presets
**Micro Parallax Product Hero**
{% code overflow="wrap" %}
```
Slow cinematic parallax push toward the product in the source image. The camera moves forward slightly while drifting a few degrees to the right, creating subtle foreground/background separation. Preserve the exact product shape, logo placement, material finish, color, and background styling. Add only gentle light shimmer and shallow depth-of-field breathing. No new objects, no logo distortion, no warping, no scene replacement.
```
{% endcode %}
Why it works: it asks for a small motion with clear preservation rules. Use it when a normal dolly-in feels too plain but you do not want an aggressive camera move.
**Luxury Tabletop Turn**
{% code overflow="wrap" %}
```
Elegant tabletop camera slide around the subject in the source image, like a controlled product commercial on a motion-control rig. Camera glides from front-left to front-right while staying close to the original framing. The subject remains the same object with identical materials, markings, colors, and proportions. Preserve the tabletop, props, and background. Add subtle specular highlights as the camera moves. No extra products, no melted edges, no label changes, no shaky handheld motion.
```
{% endcode %}
Why it works: it defines the camera rig, direction, and product-preservation needs.
**Atmospheric Establishing Drift**
Field
Value
Recommended duration
8s
Best for
Environments, concept art, landscapes, cinematic scene openers
{% code overflow="wrap" %}
```
Slow atmospheric establishing shot from the source image. Camera floats forward gently through the scene with very subtle handheld drift, revealing depth in the foreground and background. Preserve the same architecture, terrain, lighting direction, color palette, and composition. Add natural atmosphere: drifting haze, distant light flicker, slight movement in leaves or fabric if present. No new buildings, no geography changes, no warped horizon, no fast camera movement.
```
{% endcode %}
Why it works: it gives the model permission to animate atmosphere while protecting the world layout.
**Documentary Handheld Portrait**
Field
Value
Recommended duration
5s
Best for
Character portraits, gritty realism, interviews, behind-the-scenes looks
Intimate documentary handheld camera observing the person in the source image. Add natural micro-movement, slight breathing in the frame, and subtle environmental motion. The person keeps the same face, expression, hairstyle, clothing, pose, and identity. Preserve the original background and lighting. The movement should feel human and grounded, not artificial. No face morphing, no new people, no sudden zoom, no scene redesign.
Why it works: it uses handheld as texture, not chaos.
**Slow Suspense Push**
Field
Value
Recommended duration
6s
Best for
Horror, thriller, dramatic reveals, tension beats
{% code overflow="wrap" %}
```
Very slow suspenseful camera push toward the subject in the source image. The camera advances almost imperceptibly while the background feels slightly heavier and more compressed. Keep the subject's identity, clothing, pose, and facial structure unchanged. Preserve the same room and lighting, but allow subtle shadow deepening and faint atmospheric movement. No jump scare, no new figure, no face change, no background melting, no fast zoom.
```
{% endcode %}
Why it works: it creates tone without asking the model to invent a new event.
***
### AI Transitions Custom Presets
AI Transitions custom presets are for bridging a start frame and an end frame.
Use them when you want a transition style that is not covered by the built-ins, or when you have a very specific brand or editorial language.
#### How to Write Transition Prompts
A good custom transition prompt should include:
* **Start anchor**: how the shot begins
* **Transition mechanism**: what physically or optically hides the change
* **End anchor**: how the shot resolves
* **Camera behavior**: locked, push, pan, whip, fly-through, rack focus, etc.
* **Quality lock**: what must stay stable and what should not appear
Good structure:
Anchor on the start frame. First, [transition begins]. Then [the change is hidden or physically transforms]. Finally, the shot resolves into the end frame. Keep it one continuous shot. [Quality lock instructions.]
Avoid:
* "Make a cool transition"
* Asking for a normal cross-dissolve
* Ignoring the end frame
* Asking for multiple transition mechanisms at once
* Prompts that require text or logos to stay perfectly legible
***
#### Negative Prompts for Custom Transitions
AI Transitions custom presets include an optional negative prompt field.
Use it when the same artifact keeps appearing:
* ghosting
* double exposure
* visible cuts
* flicker
* warped faces
* melted hands
* fake overlays
* camera shake
* unstable background
* text artifacts
* watermarks
Do not use the negative prompt as a second creative prompt. Keep it short and failure-focused.
Good negative prompt:
{% code overflow="wrap" %}
```
ghosting, double exposure, visible cut, hard cut, flicker, warped face, extra limbs, text artifacts, watermark, unstable background
```
{% endcode %}
Weak negative prompt:
```
make it cinematic, add smoke, add lens flare, use better lighting
```
#### AI Transitions Example Presets
**Glass Reflection Flip**
Field
Value
Recommended duration
5s
Best for
Music videos, nightlife, city edits, creator transitions
Prompt:
{% code overflow="wrap" %}
```
Anchor on the start frame for a brief hold. First, thin neon light trails begin tracing the brightest edges of the scene. Then the camera moves into a fast lateral glide as the trails stretch across the frame, fully obscuring the image with directional motion blur and glowing color streaks. Finally, the blur clears and the shot resolves into the end frame with the same direction of motion. One continuous camera move, no visible cut, no cross-dissolve. The light trails are physical streaks in space, not a flat overlay.
```
{% endcode %}
Negative prompt:
visible cut, hard cut, cross dissolve, ghosting, double exposure, static overlay, flicker, warped faces, text artifacts, watermark
Anchor on the start frame with a natural handheld camera feel. First, a dark foreground shape passes very close to the lens from left to right, filling the frame with soft motion blur. Then the obstruction completely covers the image for a fraction of a second, hiding the scene change. Finally, the obstruction clears and reveals the end frame as if the camera has continued moving through the same space. The transition is motivated by the object passing the lens, not by a dissolve.
Nostalgia, fashion, music videos, analog title sequences
Prompt:
{% code overflow="wrap" %}
```
Anchor on the start frame like a locked film shot. First, warm amber exposure blooms at the edge of frame, like real film catching light. Then the burn expands organically with grain, halation, and soft overexposure, briefly washing the image into bright analog texture. Finally, the exposure rolls away and reveals the end frame beneath it. The transition should feel like practical film chemistry and light leak, not a digital wipe. Maintain one continuous shot with natural film grain and no hard cut.
```
{% endcode %}
Negative prompt:
digital wipe, hard cut, harsh edge, fake overlay, posterized color, text artifacts, watermark, flicker, unstable subject
Prompt:
{% code overflow="wrap" %}
```
Anchor on the start frame. First, the camera eases toward a reflective glass surface or glossy highlight in the scene. Then the reflection expands and bends across the frame, turning the image into a smooth mirrored distortion that briefly fills the lens. Finally, the reflection settles and resolves into the end frame, as if the camera has passed through the reflective surface. The motion is elegant, glossy, and controlled. No visible cut. No cross-dissolve. Keep the camera movement smooth and premium.
```
{% endcode %}
Negative prompt:
Art channels, title sequences, poetic scene changes, stylized reels
Prompt:
{% code overflow="wrap" %}
```
Anchor on the start frame with a locked camera. First, dark liquid ink begins spreading from the shadow areas of the image, following natural contours and edges. Then the ink blooms outward in soft tendrils, covering the frame with organic fluid motion while hints of the end frame begin appearing inside the wash. Finally, the ink recedes and resolves cleanly into the end frame. The effect is physical liquid pigment moving across the image, not a flat dissolve. Maintain stable composition and one continuous transformation.
```
{% endcode %}
Negative prompt:
{% code overflow="wrap" %}
```
flat dissolve, hard cut, digital smear, muddy blur, ghosting, double exposure, unstable subject, warped face, extra limbs, text artifacts
```
{% endcode %}
**Practical Flash Pop**
Field
Value
Recommended duration
3s
Best for
Fast social edits, camera flashes, event videos, fashion reels
Prompt:
{% code overflow="wrap" %}
```
Anchor on the start frame for a half-beat. First, a practical camera flash fires from within the scene, rapidly blooming to white with a crisp photographic strobe feel. Then the flash overexposes the entire frame for a split second, hiding the scene change. Finally, exposure settles into the end frame with a tiny afterimage and natural lens recovery. The transition is fast, clean, and motivated by a real camera flash. No visible cut, no slow dissolve.
```
{% endcode %}
Negative prompt:
***
#### Best Practices
**Name presets by the result**
Use names like "Micro Parallax Product Hero," "Film Burn Reveal," or "Lens Pass Obstruction." Avoid names like "Test 1" or "Cool Move."
**Use thumbnails for visual memory**
A thumbnail is optional, but it makes custom cards easier to recognize later. Use a small image that represents the motion or transition style.
**Keep prompts complete**
Because custom prompts run directly, include the camera behavior, preservation rules, and quality lock inside the prompt itself.
**Save only reusable ideas**
If a prompt only works for one project, keep it as a one-off. Custom presets are better for moves and transitions you expect to use again.
**Start from built-in language**
If you like a built-in style but need a variation, read the built-in page for that workflow and borrow its structure: anchor, movement, physical transition, quality lock.
**Do not overload a preset**
One preset should do one job. "Whip pan, smoke reveal, light trails, film burn, and morph" is too much for a reusable card.
***
#### Troubleshooting
**My custom preset ignores the source image more than a built-in preset**\
Built-ins use workflow-specific analysis and prompt construction. Custom presets run directly, so add stronger preservation language: "preserve the subject, clothing, pose, lighting, background, and composition."
**My Motion Director custom preset does not use my one-off action notes**\
Put the full reusable instruction in the custom template prompt. Custom Motion Director presets are designed as complete recipes.
**My transition custom preset does not adapt to the exact subject**\
Write the prompt around "the start frame," "the end frame," "the subject shown in the first frame," and "the final frame" instead of relying on automatic placeholder replacement.
**The same artifact appears every time in a custom transition**\
Add a short negative prompt that names that failure mode: ghosting, double exposure, hard cut, flicker, warped face, unstable background, etc.
**The custom card is hard to find later**\
Rename it with a result-focused name and add a thumbnail.
**I accidentally made a bad preset**\
Open the picker, click the edit icon on the custom card, then revise or delete it.
***
**Next:** Use Motion Director to build custom camera movement recipes, or use AI Transitions to turn custom transition prompts into reusable bridge styles.
# How to Generate B-Roll Inside Premiere Pro with AI
Generate cinematic B-roll, establishing shots, and visual coverage directly inside Premiere Pro — powered by Seedance 2 and Kling 3.0. Access the world's most advanced AI video models on a pay-per-cli
### The Problem
B-roll gaps are one of the most common blockers in a video edit. A missing establishing shot. A cutaway that doesn't exist. An atmosphere clip you never filmed. The traditional fix — stock footage — either costs a subscription, looks generic, or doesn't match the mood of your edit at all.
The alternative is AI video generation. But every service that offered it required a separate subscription, a separate browser tab, and a separate export/import process every time you needed a clip. The context switching added friction to what should be a 60-second task.
***
### The Solution: Two Video Model Giants, Inside Premiere Pro
The AI video generation landscape has a new tier. Two models have established themselves as the best in class for cinematic content — and both are now accessible directly inside Premiere Pro through Chat Video Pro, on a pay-per-clip basis with no subscription required.
**Seedance 2** and **Kling 3.0** are the models editors are calling game-changers. They represent different strengths, and understanding which to reach for first is the key to fast, high-quality b-roll generation.
***
### Seedance 2 — Cinematic Environments at 1080p
Seedance 2 (by ByteDance) is the model the industry is calling a landmark in cinematic generation for non-human content. For environments, landscapes, architecture, objects, abstract scenes, and establishing shots, it produces footage that is visually indistinguishable from professional cinematography — at native 1080p resolution, up to 15 seconds, with ambient audio included.
**What it excels at:**
* **Establishing shots** — cityscapes, aerial views, wide landscapes, architecture
* **Nature and environment** — forests, coastlines, weather, golden hour, golden fog
* **Abstract and conceptual** — motion graphics-style movement, abstract light, geometric scenes
* **Product and object b-roll** — objects on surfaces, product reveals, detail shots
* **Atmosphere** — rain, mist, smoke, neon reflections, interiors without people
* **Cinematic camera movement** — slow push-ins, drifting crane shots, natural handheld
**Native audio.** Seedance 2 generates ambient sound automatically — wind, city hum, rain, footsteps. For many b-roll shots, the clip arrives ready to drop in with sound already there.
**15-second duration.** Seedance 2 runs up to 15 seconds with native audio, giving you room to trim and find the perfect moment.
**The caveat:** Human faces and close-up people content are not Seedance 2's strongest territory. For b-roll with people in the scene, Kling 3.0 is the better choice.
**Prompt examples:**
```
"Slow aerial push over a foggy mountain valley at sunrise. Golden light through the mist.
Cinematic, 4K feel."
"Close-up of rain hitting a puddle on a city street at night, neon reflections.
Slow motion."
"Modern office interior, late afternoon, empty desks, window light streaming in.
Wide shot, slow drift."
"Product shot — sleek laptop on a minimal wooden desk, subtle focus pull.
Clean and professional."
```
***
### Kling 3.0 — The Human Subject Champion
Kling 3.0 (by Kuaishou) is the model you reach for when there are people in your b-roll. Realistic faces, natural body movement, human interactions, lifestyle scenes, street-level footage with subjects — this is where Kling 3.0 outperforms the field.
**What it excels at:**
* **People-focused b-roll** — walking, working, interacting, reacting
* **Lifestyle content** — someone at a coffee shop, working at a computer, in conversation
* **Professional contexts** — office environments with people, presentations, handshakes
* **Emotional close-ups** — face reactions, expressions, contemplation
* **Action and movement** — sports, running, physical activity
* **Social scenes** — groups of people, crowds, candid-style footage
**Kling 3.0 Pro** delivers the highest V3 quality — more detail, better motion, stronger consistency. Use it for the takes that matter.
**Kling 3.0 Standard** is the fast-iteration version. Generate 5 variations quickly, pick the best one, then refine with Pro.
**Native audio** on all Kling 3.0 models, including ambient sound and basic sound effects.
**Prompt examples:**
```
"Young professional woman walking through a modern office hallway, confident,
natural lighting, shallow depth of field."
"Close-up of hands typing on a laptop, coffee cup in background, soft focus.
Warm natural light."
"Two people shaking hands in a bright business meeting room.
Professional, authentic."
"Person running on a city street at dawn, motion blur, cinematic."
```
***
### The Decision: Which Model to Reach For
| Scene type | Best model |
| ------------------------------------ | --------------------------------- |
| Landscape, environment, nature | Seedance 2 |
| Architecture, cityscape, aerial | Seedance 2 |
| Abstract, conceptual, atmospheric | Seedance 2 |
| Product or object b-roll | Seedance 2 |
| People walking, working, reacting | Kling 3.0 Pro |
| Lifestyle and human interaction | Kling 3.0 Pro |
| Action, sports, physical movement | Kling 3.0 Pro or Hailuo 03 |
| High-speed iteration on people shots | Kling 3.0 Standard |
| Transitions between two keyframes | Seedance 2 or Kling O3 Transition |
***
### Why This Changes the B-Roll Workflow
Before Chat Video Pro, accessing Kling required a subscription ($66/month for 660 credits). Runway was $35/month. Pika was $35/month. And none of them lived inside Premiere Pro — every clip required a browser tab, a prompt, a download, and a re-import.
With Chat Video Pro:
* **No additional subscriptions.** Pay per generation, no monthly fees for the video models themselves
* **No context switching.** Generate from inside your edit, drop directly into your Library and timeline
* **Multiple models at once.** Switch between Seedance 2, Kling 3.0, Veo 3.1, and others in the same session without logging in or out of anything
* **Frame Capture as context.** Grab a frame from your timeline and hand it to the AI as a style reference — matching your footage's color, lighting, and composition in a way no stock footage library can
***
### The Workflow
#### 1. Identify the Gap
While cutting, note what's missing — a missing establishing shot, a cutaway, an atmosphere clip, a concept visual. Note the composition you need (wide, close-up, POV), the mood, and whether there are people in it.
#### 2. Choose Your Model
**People in the shot?** → Kling 3.0 Pro\
**No people — environment, landscape, object?** → Seedance 2\
**Dialogue or speaking character?** → Veo 3.1\
**Action, sports, fast movement?** → Hailuo 03
#### 3. Capture a Frame (Optional)
Position your playhead on a nearby shot in your sequence that has the right mood or visual style. Click **Frame Capture**. The frame attaches to your composer as visual reference — the AI can match your footage's lighting, color temperature, and composition.
#### 4. Write Your Prompt
A strong b-roll prompt has four elements:
```
[Subject/scene] + [Camera movement] + [Lighting/mood] + [Style]
```
Example:
```
"Aerial drift over a coastal town at golden hour. Slow, gentle movement.
Warm evening light. Cinematic, Seedance 2."
```
Describe what you see, not what you want it to mean. "Rainy street at night with neon reflections" is better than "moody urban atmosphere."
#### 5. Generate and Import
Generated clips land in your Chat Video Pro Library. Preview them in the panel, pick the best take, and drag it directly into your Premiere Pro timeline. If the first result isn't quite right, adjust the prompt and generate again — the whole loop takes 60–90 seconds.
***
### Beyond Basic B-Roll
**Reference Mode — consistent subjects across multiple clips.** Kling O3 Reference accepts up to 7 reference images and maintains character or object consistency across generations. For a branded series where b-roll needs to feature the same product or person, generate multiple clips that stay visually consistent.
Seedance 2 Reference accepts up to 9 reference images and 3 audio reference files — useful when you want ambient sound from a specific sonic environment.
**Transition Mode — generate the moment between two frames.** Attach two screenshots from your timeline as start and end frames. Seedance 2 and Kling O3 Transition generate the cinematic motion connecting them. Useful for covering a scene change, revealing a location, or creating a visual bridge between shots.
**Kling Multi-Cam — generate alternative angles from existing footage.** Kling O3 Multi-Cam takes an existing clip and generates new camera perspectives from it. Already have a wide shot? Generate a close-up, a reverse angle, or a different focal length from the same moment — without a second camera.
***
### Tips for Best Results
**Match your footage, not stock.**\
Use Frame Capture to anchor your generation to your actual edit. Hand Seedance 2 or Kling a frame from your sequence, describe the b-roll you need, and the result will naturally match your video's visual language.
**Generate more than you need.**\
Request 3–4 takes with slightly different prompts. AI generation has variance; give yourself options so you're choosing the best take, not hoping the first one works.
**Let Seedance 2 generate the audio.**\
For environment and atmosphere shots, let the native audio run. Seedance 2's ambient sound generation is strong enough that you may not need to replace it — just blend the levels in your mix.
**Use Kling 3.0 Standard for fast iteration.**\
If you're not sure about the composition or framing, use Standard to iterate quickly on 3–5 variations. Once you find the right shot, generate a final version with Pro for the highest quality take.
**Describe camera movement specifically.**\
"Slow push-in" and "FPV drone shot" and "static wide" produce very different results. Model both Kling and Seedance well to cinematic camera language — use it.
***
### Next Steps
* [**Supported Video Models**](/features/video-generation/supported-video-models) — Full specs, duration limits, and audio capabilities for every model
* [**Reference Mode**](/features/video-generation/reference-mode) — Keep subjects and characters consistent across multiple generations
* [**Transition Mode**](/features/video-generation/transition-mode) — Generate cinematic motion between two keyframes
* [**Frame Capture**](/getting-started/interface-overview/frame-capture-button) — Capture frames from your timeline as generation context
***
**Related Workflows:**
* [AI VFX Inside Premiere Pro](/workflows/how-to-create-ai-powered-visual-effects-in-premiere-pro)
* [Story Editing with AI in Premiere Pro](/workflows/how-to-cut-videos-faster-with-ai-assisted-story-editing-in-premiere-pro)
* [AI Thumbnails in Premiere Pro](/workflows/how-to-generate-high-ctr-thumbnails-inside-premiere-pro-with-ai)
# How to Create AI-Powered Visual Effects in Premiere Pro
Add visual effects, transform scenes, and modify footage using AI—all within Premiere Pro. Test VFX ideas, apply effects, and iterate without leaving your timeline. No need for After Effects.
### The Problem
Video editors and content creators often break their creative flow when they need to test visual effects ideas. The typical workflow involves exporting clips, opening After Effects or external compositing tools, applying effects, rendering, and re-importing—only to discover the effect doesn't quite work or needs adjustment. This back-and-forth kills momentum, especially when you're experimenting with multiple VFX ideas or need to iterate quickly.
The problem intensifies for creators working on tight deadlines, social media content, or projects where you need to test multiple effect variations. Each round trip between Premiere Pro and external tools adds time, breaks focus, and makes experimentation feel costly rather than creative.
***
### The Solution
A streamlined workflow keeps visual effects experimentation inside your editing environment, allowing you to test ideas, apply effects, and refine results without leaving your timeline. Instead of exporting and switching applications, you can modify footage directly, preview changes in context, and iterate on visual effects while maintaining your editing flow.
This approach treats VFX as part of the creative exploration process rather than a separate, time-consuming step. You can experiment freely, compare variations, and make decisions based on how effects look within your actual edit—not in isolation.
***
### How Chat Video Pro Implements This
Chat Video Pro brings AI-powered visual effects directly into Adobe Premiere Pro through the Video Canvas Editor. You can:
* [**Import clips** ](/getting-started/interface-overview/import-clip-button)**from your timeline** - Use the Import Clip button to bring footage directly into the editor, or drag and drop from your Library
* **Apply effects instantly** - Select from AI models like Kling VFX to add rain, fire, lighting changes, weather effects, and atmospheric modifications
* **Preview in context** - See effects applied to your actual footage before committing
* **Iterate quickly** - Generate multiple variations, compare before/after, and refine until you get the look you want
* **Import back seamlessly** - Edited footage appears in your Library and can be dragged directly back into your Premiere Pro sequence
The entire VFX workflow happens inside Premiere Pro, so you never lose context or break your editing rhythm. Effects become part of your creative process, not a separate production step.
***
### The Workflow Steps
#### 1. Identify the Effect You Want
While editing, identify where visual effects would enhance your footage:
* Weather effects (rain, snow, fog)
* Lighting changes (day to night, golden hour, dramatic lighting)
* Atmospheric modifications (mood, tone, style)
* Scene transformations (time of day, season changes)
* Style transfers (cinematic looks, artistic treatments)
#### 2. Import Your Clip
* Select the clip in your Premiere Pro timeline
* Click the Import Clip button in Chat Video Pro
* Or drag and drop the clip from your timeline or Library
* The clip appears in the composer ready for editing
#### 3. Open Video Canvas Editor
* Click the "Edit" button on the video thumbnail
* Video Canvas Editor opens in full-screen
* Your clip loads and is ready for effects
#### 4. Select Kling VFX
* Choose "Kling VFX" from the model selector
* The interface adapts to show VFX-specific controls
* You can now add effects to your footage
#### 5. Apply Your Effect
**Simple method (video only):**
* Enter a prompt describing the effect you want
* Example: "Add rain to the scene" or "Make it nighttime with dramatic lighting"
* Click "Edit Video" to apply
**Power method (video + reference image):**
* Create a reference image showing the desired effect (using Nano Banana or another tool)
* Add the reference image to the editor
* Reference it in your prompt: "Make the scene look like @Image1"
* This shows the model exactly what you want
#### 6. Review and Refine
* Use the Before/After toggle to compare results
* If the effect needs adjustment, modify your prompt and regenerate
* Generate multiple variations to find the best look
* All without leaving Premiere Pro
#### 7. Import Back to Timeline
* Click "Done" when satisfied with the effect
* The edited clip appears in your Library
* Drag and drop it back into your Premiere Pro sequence
* Replace the original clip or use it as a new layer
***
### When This Workflow Is Useful
This workflow is ideal for:
* **Content creators** - Adding atmospheric effects to enhance storytelling
* **Social media teams** - Creating eye-catching effects for short-form content
* **YouTube creators** - Enhancing footage with weather, lighting, or style effects
* **Documentary editors** - Modifying footage to match historical periods or moods
* **Creative experimentation** - Testing visual ideas without committing to complex compositing
* **Quick enhancements** - Adding polish to footage that needs a visual boost
* **Style consistency** - Applying consistent looks across multiple clips
***
### When This Workflow May Not Be Suitable
This workflow may not replace:
* **Complex compositing** - Multi-layer effects, advanced masking, or intricate keying that requires After Effects
* **Professional VFX pipelines** - Feature film or broadcast work requiring frame-by-frame control and specific technical requirements
* **Real-time effects** - Effects that need to be applied during live production or broadcast
* **Extremely precise control** - When you need pixel-perfect control over every aspect of the effect
* **Industry-standard workflows** - Projects requiring specific VFX software for client deliverables or team collaboration
The AI-powered effects work best as **creative enhancements** and **atmospheric modifications** rather than complex, multi-layer compositing work.
***
### Tips for Best Results
#### Effect Prompt Writing
* **Be specific about the effect**: "Add heavy rain with water pooling on the ground" vs. "rain."
* **Describe the mood**: "Dramatic nighttime scene with neon street lighting" vs. "nighttime."
* **Mention intensity**: "Light snowfall" vs. "heavy blizzard."
* **Include context**: "Golden hour lighting on a city street" vs. "golden hour"
#### The Power Workflow: Nano Banana + Kling VFX
The most effective way to use Kling VFX is to create a reference image first:
1. **Create stylized reference** - Use Nano Banana to generate an image showing the exact effect you want
* Example: "Make this scene nighttime with neon lights and wet streets."
* Example: "Transform this to a snowy winter scene with falling snow."
* Example: "Change the lighting to golden hour with warm tones"
2. **Use as reference** - Import your video, add the Nano Banana image as reference, and use `@Image1` In your prompt
* This shows the model exactly what you want
* Results are more accurate and closer to your vision
#### Matching Your Footage
* **Keep clips short** - 3-10 seconds works best for VFX processing
* **Good source footage** - Well-lit, clear footage produces better results
* **Consider composition** - Effects work best on footage with clear subjects and backgrounds
* **Test first** - Try effects on short clips before processing longer sequences
#### Integration Tips
* **Use Before/After toggle** - Always compare original and modified footage
* **Generate variations** - Create multiple versions to find the best look
* **Save to Library** - Build a collection of effect variations for reuse
* **Layer effects** - You can apply multiple effects by processing clips sequentially
***
### Common Use Cases
#### Weather Effects
Transform scenes with weather:
* "Add heavy rain to the street scene."
* "Make it snow with snow accumulating on surfaces."
* "Add fog and atmospheric mist."
* "Create a stormy sky with dramatic clouds"
#### Lighting Transformations
Change time of day and lighting:
* "Transform to nighttime with street lighting."
* "Make it golden hour with warm, soft lighting."
* "Create dramatic lighting with strong shadows."
* "Change to blue hour with cool tones"
#### Atmospheric Modifications
Alter mood and atmosphere:
* "Make the scene more cinematic with film grain."
* "Add a vintage, nostalgic look."
* "Create a futuristic, cyberpunk atmosphere."
* "Transform to a dreamy, ethereal mood."
#### Style Transfers
Apply artistic treatments:
* "Make it look like a film noir scene."
* "Apply a warm, vintage color grade."
* "Create a high-contrast, dramatic look."
* "Transform to a desaturated, moody aesthetic"
***
### Advanced Techniques
#### Character Swapping
Kling VFX can also swap characters in videos:
* Add 1-4 reference images of the character you want
* Use `@Image1`, `@Image2`, et,c. in your prompt
* Example: "Replace the person with the character from @Image1"
#### Combining Effects
You can layer effects by processing clips multiple times:
* First pass: Add weather effect
* Second pass: Modify lighting
* Third pass: Apply style treatment
* Each pass builds on the previous result
#### Reference Image Workflow
For maximum control:
1. Generate a reference image with Nano Banana showing the desired effect
2. Import video and add reference image
3. Prompt: "Make the scene match the style and atmosphere of @Image1"
4. This gives you precise control over the final look
***
### Next Steps
* [**Learn about Kling VFX**](/features/studio/kling-vfx): Kling VFX Feature Guide
* [**Understand Video Canvas Editor**](/features/video-generation/video-canvas-editor): Video Canvas Editor
* [**See how to import clips**](/getting-started/interface-overview/import-clip-button): Import Clip Button
* [**Learn about Nano Banana**](/features/image-generation): Image Generation Features
* [**Check pricing**](/getting-started/interface-overview/usage-panel): Pricing Information
***
**Related Workflows:**
* [Generate AI B-Roll Directly Inside Premiere Pro](/workflows/how-to-generate-b-roll-inside-premiere-pro-with-ai)
* [Story Editing with AI](/workflows/how-to-cut-videos-faster-with-ai-assisted-story-editing-in-premiere-pro)
* [AI Thumbnails in Premiere Pro](/workflows/how-to-generate-high-ctr-thumbnails-inside-premiere-pro-with-ai)
# How to Cut Videos Faster with AI-Assisted Story Editing in Premiere Pro
Export a transcript from Premiere Pro, attach it to Chat Video Pro, and get a complete rough cut inserted directly into your timeline — with section markers, verbatim soundbites, and exact timestamps.
### The Problem
For dialogue-heavy content — interviews, podcasts, documentary footage — the hardest part of editing isn't the cut itself. It's finding the moments. Scrubbing hours of footage to locate the best soundbites, structure a narrative arc, and deliver multiple platform versions from one shoot can take days of work before a single clip is placed.
The standard workflow compounds the problem: watch everything, take notes, repeat for every deliverable. By the time you start cutting, you've spent more time on research than editing.
> **This workflow is built for dialogue-heavy content.** Story Cutter reads spoken word from a transcript. It identifies and structures soundbites — it does not analyze b-roll, graphics, or visual-only footage.
***
### The Solution
Export a transcript from Premiere Pro's built-in Text panel, attach it to Chat Video Pro, describe your goal, and click **Insert Rough Cut**. Your entire first cut — trimmed soundbites in story order, with section markers on the timeline — appears in Premiere Pro in seconds.
No scrubbing. No manual logging. No switching apps. The AI finds the moments; you make the creative decisions.
For multi-deliverable shoots, `/batch` generates a TikTok cut, a YouTube cut, and a long-form version simultaneously from the same transcript — all in a single message.
***
### How It Works
#### 1. Export Your Transcript from Premiere Pro
Open the **Text** panel in Premiere Pro (Window → Text), transcribe your footage, then click the three dots (⋯) → **Export as transcript file** → save as a **`.json` file**.
> **Use Premiere Pro's native export.** The `.json` format carries timestamp data tied to each clip's exact position on your timeline. SRT, VTT, plain text, and exports from Descript, Rev, or Otter are not supported.
#### 2. Start Story Cutter and Attach the Transcript
Click the **Story Cutter** conversation starter in Chat Video Pro. Use the paperclip (📎) button to attach your `.json` transcript. A purple pill appears in the composer confirming the transcript is loaded.
When the pill appears, click it to link Story Cutter to the exact Premiere Pro timeline you transcribed. If your transcript file name matches your timeline name, it links automatically.
> **Don't move clips after transcribing.** Timestamps are tied to clip positions. Rearranging footage breaks the sync and requires a new transcript export.
#### 3. Attach a Script or Creative Brief (Optional)
Attach any reference document alongside your transcript — a shooting script, interview outline, client brief, or show format. Story Cutter reads it and uses it as a structural guide for soundbite selection and ordering. The more specific the brief, the more intentional the cut.
Useful documents to attach:
* Original shooting script or interview question set
* Client brief or creative direction PDF
* Show bible or episode format
* Notes from a previous cut
**Supported formats:** `.pdf`, `.docx`, `.txt`, `.md`
#### 4. Write Your Prompt
Describe the goal in one or two sentences. Tell it the platform, runtime, and what the video is about:
> "5-minute YouTube tutorial about how to set up a Premiere Pro workspace. Start with a strong hook, end with a CTA to subscribe."
> "60-second Instagram Reel from the Q\&A section. Energetic, hook-first, no interviewer questions."
For more complex briefs, use `/creative brief` to plan the structure before cutting.
#### 5. Review the Paper Cut and Insert
Story Cutter streams back a structured paper cut — verbatim quotes, `HH:MM:SS:FF` timestamps, section labels, and editor notes.
**Three ways to use the results:**
| Action | What it does |
| -------------------------- | -------------------------------------------------------------------------------------------- |
| **Click a timestamp** | Jumps your Premiere Pro playhead to that moment so you can preview it |
| **Click ↓ on a soundbite** | Inserts just that one soundbite at your playhead |
| **Insert Rough Cut** | Places the entire selection onto your timeline at once, in story order, with section markers |
**Insert Rough Cut** is the fastest path to a first cut. Scroll to the bottom of the paper cut, click the button, and the whole edit lands on your timeline — trimmed and structured — ready to refine.
#### 6. Refine in the Same Thread
Story Cutter holds the full context of your transcript and previous selections. Refine without re-explaining:
* "Cut 30 seconds from section 2 — it runs too long"
* "The opening doesn't land — find a stronger hook"
* "Replace the third soundbite, it feels repetitive"
* "Remove anything about pricing"
* "Find a cleaner close"
Each refinement produces a new paper cut with an updated Insert Rough Cut button. Keep iterating until it's right, then insert.
***
### Hard Constraints: Precision Editing
Add precise constraints at any point in the conversation:
| Constraint | Example |
| ----------------- | ------------------------------------------------ |
| Timecode range | "Only use footage between 00:05:00 and 00:45:00" |
| Required topics | "Must include the part about the funding round" |
| Excluded topics | "Skip anything about the early team" |
| Named soundbites | "Start with the line about failure" |
| Start/end moments | "Start at 00:03:22, end with the outro" |
| Tone | "Keep the tone measured — no hype" |
Constraints stack. You can combine multiple in a single message and they apply directly to the next paper cut.
***
### /batch — Multiple Deliverables, One Pass
For shoots where you need TikTok, YouTube, and long-form versions, `/batch` generates all of them from one transcript in a single message.
```
/batch
- 60-second TikTok Reel, energetic hook, focus on the failure story
- 3-minute YouTube, Problem → Credibility → Steps → CTA
- 90-second Instagram Reel, emotional tone
```
Each task produces its own paper cut and Insert Rough Cut button. Results appear in sequence in the same thread.
**Using `/batch` with a brief:** If you attach a document that enumerates multiple version goals ("Version 1: founder story. Version 2: product demo. Version 3: customer outcome"), `/batch` treats each as a separate cut target.
> `/batch` processes one transcript at a time — it produces multiple cuts from one source, not from multiple separate transcript files.
***
### Common Use Cases
**Long-form interview to multi-platform** Upload the transcript, run `/batch` for TikTok + YouTube Shorts + long-form in one pass. Three paper cuts, three Insert Rough Cut buttons. Done.
**Script-guided documentary** Attach your scene breakdown or narrative outline alongside the transcript. Story Cutter surfaces soundbites that match your intended arc and structures the cut to follow your document.
**Branded content with a client brief** Write the creative brief as a PDF — include tone, required topics, structure, and anything to avoid. Attach it with the transcript. The cut follows the brief without rounds of back-and-forth.
**Large transcript, fast selects** Run `/select pass` to surface every strong moment categorized by type — hooks, value moments, emotional peaks, CTAs. Then use individual ↓ inserts to hand-pick the best ones into your edit.
***
### Slash Commands
| Command | What it does |
| ------------------- | -------------------------------------------------------- |
| `/batch` | Multiple deliverables from one transcript in one pass |
| `/social clip` | 60-second social-optimized cut |
| `/select pass` | Full transcript scan — surfaces best moments by category |
| `/top 5 soundbites` | Five strongest moments with platform recommendations |
| `/creative brief` | Plan-first flow — outlines the cut before generating it |
| `/new video` | Clears the transcript and resets for a new project |
***
### Story Structures Available
Request any of these by name, or let the AI choose based on your platform and content.
| Platform | Structure |
| ------------------------------- | -------------------------------------------------------------- |
| TikTok / Reels / Shorts | Hook → Payoff → Proof → CTA |
| YouTube (2–3 min) | Problem → Credibility → Steps → Result → CTA |
| Tutorial / Educational (4+ min) | Cold-Open → Context → Rising Tension → Resolution → Reflection |
| Documentary | Three-Act: Setup → Confrontation → Resolution |
***
### Next Steps
* [**Story Cutter Assistant Guide**](/conversation-starters/story-cutter-assistant) — Detailed setup, multicam configuration, Insert Rough Cut behavior, and all slash commands
* [**Usage and Pricing**](/getting-started/interface-overview/usage-panel) — Understand how credits work
***
**Related Workflows:**
* [Generate AI B-Roll Directly Inside Premiere Pro](/workflows/how-to-generate-b-roll-inside-premiere-pro-with-ai)
* [AI VFX Inside Premiere Pro](/workflows/how-to-create-ai-powered-visual-effects-in-premiere-pro)
* [AI Thumbnails in Premiere Pro](/workflows/how-to-generate-high-ctr-thumbnails-inside-premiere-pro-with-ai)
# How to Generate High-CTR Thumbnails Inside Premiere Pro with AI
Generate YouTube thumbnails, social media graphics, and visual assets directly inside Premiere Pro — powered by GPT Image 2. Crisp text overlays, complex graphic compositions, and multi-format exports
### The Problem
Thumbnail creation is where editing flow goes to die. The typical workflow: pause your edit, open Photoshop or Canva, design the thumbnail, export it, import it back, realize it doesn't match your video's feel, iterate. By the time you have something usable, you've broken your editing rhythm and spent 30–60 minutes on what should be a 5-minute task.
For creators running multiple videos or needing A/B variations, this context switching compounds fast. The design work is separate from the creative work — and the separation is the problem.
***
### The Solution: GPT Image 2, Inside Premiere Pro
With the release of GPT Image 2, AI thumbnail generation has crossed a threshold that changes the workflow entirely. Two capabilities matter most for thumbnails:
**Accurate text rendering.** The historic weakness of AI image generation for thumbnails was text — garbled letters, broken words, and unusable overlays. GPT Image 2 renders clean, legible text directly in the image. Bold type, title text, numbers, and calls-to-action are generated accurately. You can prompt for the exact text you want and get it.
**Complex graphic compositions.** GPT Image 2 handles multi-element scenes — a person, a product, bold text, a colored background, and a graphic accent — as a single cohesive image. The kind of composition that used to require Photoshop layer work is now a prompt.
Combined with Chat Video Pro's Frame Capture, Thumbnail Mode, and Canvas Editor, the full thumbnail workflow now lives inside Premiere Pro. Capture a frame, generate, refine, export — without leaving your project.
***
### How It Works
#### 1. Capture a Frame from Your Edit (Optional but Recommended)
Position your playhead on a strong moment in your sequence — a reaction shot, a product reveal, a key visual. Click the **Frame Capture** button in Chat Video Pro. The frame attaches to your composer as image context.
This gives GPT Image 2 your video's actual visual style — lighting, color palette, your subject's face — so the thumbnail feels like it belongs to the video, not like a generic stock image.
#### 2. Enable Thumbnail Mode
In Chat Video Pro:
* Enable **Generate Media** → select **Image**
* Select **GPT Image 2** as the model
* Open **Thumbnail Mode** settings and choose:
* **On** — Full thumbnail optimization, your prompt enhanced with platform best practices
* **On with Blueprints** — Everything from "On" plus high-performing reference thumbnails attached for style inspiration
**On with Blueprints** + GPT Image 2 is the recommended combination for professional thumbnails. Blueprints provide composition reference; GPT Image 2's rendering quality executes it.
#### 3. Write Your Prompt — Include the Exact Text You Want
GPT Image 2 renders text accurately, so include it in your prompt:
> "YouTube thumbnail. Person reacting to a shocking result. Bold white text on the left: 'I Tried This for 30 Days'. Red highlight on the text. Clean dark background with subtle gradient."
> "YouTube thumbnail, 16:9. Split composition: messy desk on left, clean minimal workspace on right. Bold text in center: 'BEFORE vs AFTER'. Bright, high-contrast."
> "Tech product thumbnail. Close-up of hands holding a phone with a glowing screen. Large text overlay: '#1 Productivity App'. Clean professional look."
The more specific your text, the more accurately it renders. Spell it out exactly as you want it to appear.
#### 4. Generate Multiple Variations
Ask for 2–4 thumbnails in one generation. Thumbnail Mode creates genuinely different variations — different compositions, color schemes, text treatments, and visual approaches — from the same concept. Review them side by side and pick the one that works, or take the strongest elements from multiple.
```
"Generate 3 thumbnail variations with different text treatments and
color approaches. All should use the captured frame as style reference."
```
#### 5. Refine with the Canvas Editor
If a generated thumbnail is 90% right and needs polish — different text, an element swapped, background adjusted — open the **Canvas Editor** without leaving Premiere Pro. Describe changes in plain language or use the Canvas tools directly. GPT Image 2 supports up to 4 input images in edit mode, so you can combine elements from multiple generations.
#### 6. Adapt to Other Platforms
Re-attach the final thumbnail to the composer, change the aspect ratio selector, and prompt:
> "Optimize this for vertical Instagram (9:16) — recompose so the subject and text fit the frame."
One thumbnail concept, multiple platform formats — in a single session.
***
### Why GPT Image 2 Specifically for Thumbnails
Capability
Why it matters for thumbnails
Accurate text rendering
Bold title text, numbers, and CTAs are generated cleanly — no more garbled overlays
Complex multi-element scenes
Person + product + text + background in one cohesive image
Up to 1792×1792 resolution
Native high resolution for YouTube's recommended 1280×720 minimum
8 native aspect ratios
16:9 for YouTube, 9:16 for Reels, 1:1 for Instagram — no cropping
Up to 4 input images
Feed your captured frame + style references for grounded generation
Canvas Editor support
Multi-layer editing and targeted modifications without Photoshop
***
### What to Include in Your Prompt
**For highest CTR:**
* Describe the **emotion** first — "surprised", "confident", "shocked", "determined"
* Specify the **exact text** to appear — GPT Image 2 will render it accurately
* Describe the **composition** — person left, text right; before/after split; product in foreground
* Mention the **color approach** — "high contrast", "bright primary colors", "dark cinematic"
* Reference your **niche** if relevant — the Thumbnail Mode system applies niche-specific best practices
**Example prompt anatomy:**
```
[Emotion/subject] + [Text overlay (exact wording)] + [Composition] + [Color/style]
```
```
"Excited person looking at a laptop screen. Bold text overlay: 'This Changed Everything'.
Split-screen composition. Bright blue and white palette. YouTube thumbnail style."
```
***
### When to Use Each Thumbnail Mode Option
Scenario
Best Option
Quick concept test
Thumbnail Mode: On, GPT Image 2
Professional deliverable
Thumbnail Mode: On with Blueprints, GPT Image 2
Style needs to match your video
Frame Capture + Thumbnail Mode: On, GPT Image 2
Need multiple A/B variations fast
Generate 3–4 with any mode
Final polish after generation
Canvas Editor, Thumbnail Mode: Off
Creating a graphic (not a thumbnail)
Thumbnail Mode: Off, GPT Image 2 or Flux 2 Max
***
### Common Use Cases
**YouTube channel with consistent style** Capture a frame from every video, attach it to your thumbnail prompt, and generate with Blueprints enabled. The Frame Capture anchors the style to your actual footage; the blueprints keep the composition in line with what works on the platform. Consistent look with minimal manual effort.
**A/B testing for a new format** Generate 4 different variations in one pass — different text treatments, different emotions, different compositions. Upload all four to YouTube, split-test, and let the data tell you which approach resonates. The cost is a single generation session.
**Fast turnaround for a multi-video project** For every video in a batch, capture the best frame, write a one-sentence thumbnail brief, generate 2 variations. With GPT Image 2, you don't need to manually add text in Photoshop afterward — prompt the text directly and it renders in the image.
**Social media graphic** Need a title card, quote graphic, or announcement image? Generate with Thumbnail Mode off and prompt the exact text and composition you want. GPT Image 2's text accuracy makes it viable for graphics where text precision matters.
***
### Tips for Best Results
* **Include the exact text you want rendered.** GPT Image 2 handles it accurately — take advantage of this.
* **Describe the emotion, not just the visual.** "Excited person discovering a solution" produces better results than "person at a desk."
* **Use Frame Capture for consistency.** Your video's actual lighting and color palette, injected directly into the generation context.
* **Start with Blueprints.** The composition patterns in high-performing thumbnails are loaded automatically — let them do the heavy lifting.
* **Use Canvas Editor for the final 10%.** Not every thumbnail needs it, but when you need to swap one element or adjust a color, it's faster than re-generating from scratch.
***
### Next Steps
* [**Thumbnail Mode Feature Guide**](/features/image-generation/thumbnail-mode) — Mode options, supported models, multi-thumbnail generation
* [**Canvas Editor**](/features/image-generation/canvas-editor) — Multi-layer editing and targeted modifications
* [**Frame Capture**](/getting-started/interface-overview/frame-capture-button) — How to capture frames from your timeline
* [**GPT Image 2 Overview**](/features/image-generation/supported-image-models) — Full model specs, resolution, aspect ratios, input limits
***
**Related Workflows:**
* [Generate AI B-Roll Directly Inside Premiere Pro](/workflows/how-to-generate-b-roll-inside-premiere-pro-with-ai)
* [Story Editing with AI in Premiere Pro](/workflows/how-to-cut-videos-faster-with-ai-assisted-story-editing-in-premiere-pro)
* [Stay in Creative Flow While Editing](/workflows/how-to-stay-in-creative-flow-while-editing-with-ai-tools-in-premiere-pro)
# How to Stay in Creative Flow While Editing with AI Tools in Premiere Pro
Discover ways to maintain creative momentum in Premiere Pro using Chat Video Pro.
### The Problem
Video editors and content creators lose creative momentum every time they leave Premiere Pro to search for answers, find tutorials, generate graphics, remove backgrounds, or create publishing assets. Each context switch—opening a browser to search, switching to Photoshop for graphics, using Canva for thumbnails, or leaving to find help—breaks your flow state and makes the creative process feel fragmented.
The problem intensifies when you're in the middle of editing and hit a roadblock. You need to know how to do something in Premiere Pro, but leaving to search breaks your momentum. You need a graphic or icon, but switching to design tools interrupts your flow. You need to remove a background, but opening external tools adds friction. Each interruption compounds, turning what should be a smooth creative process into a series of disconnected tasks.
This constant context switching is especially damaging for flow state—that mental state where you're fully immersed in your work, ideas flow naturally, and time seems to disappear. Every time you leave your editing environment, you break that state and have to rebuild it, wasting mental energy and creative momentum.
***
### The Solution
A unified workflow keeps all your AI-powered tools, assistance, and creative capabilities inside your editing environment, allowing you to stay in flow state throughout your entire creative process. Instead of switching between applications, you can get answers, generate assets, remove backgrounds, control with voice, and create publishing materials—all without leaving Premiere Pro.
This approach treats video editing as a continuous creative flow rather than a series of disconnected tasks. You can solve problems, create assets, and complete your entire publishing workflow while maintaining your editing rhythm and creative momentum.
***
### How Chat Video Pro Implements This
Chat Video Pro brings AI assistance, generation, and publishing tools directly into Adobe Premiere Pro, creating a unified workspace that keeps you in flow. Here are five key ways it maintains your flow state:
***
### 1. Answer Your Questions Without Leaving Premiere Pro
**The Flow Killer:** Stopping your edit to search Google, watch YouTube tutorials, or browse forums for Premiere Pro help.
**The Flow Solution:** Get instant answers, mini-lessons, and helpful resources directly inside Premiere Pro.
Chat Video Pro acts as the built-in assistant that never requires you to leave your project:
* **Ask questions naturally** - Type or speak your question in plain English
* **Get instant answers** - The AI searches its knowledge base and provides immediate help
* **Receive mini-lessons** - Get step-by-step explanations of Premiere Pro features
* **Find tutorials and resources** - The AI finds and links to helpful tutorials, documentation, and web resources
* [**Premiere Pro Guru** ](/conversation-starters/premiere-pro-guru)- A specialized assistant trained specifically on Premiere Pro workflows, troubleshooting, and best practices
**Workflow:**
1. While editing, you encounter something you don't know how to do
2. Ask Chat Video Pro: "How do I create a nested sequence?" or "What's the best way to sync audio?"
3. Get an immediate answer with step-by-step instructions
4. Optionally ask for more detail or related techniques
5. Continue editing without breaking the flow
**Example Questions:**
* "How do I create a nested sequence?"
* "What's the best way to sync audio in Premiere Pro?"
* "Show me how to use the Lumetri Color panel."
* "Find me a tutorial on color grading."
* "What keyboard shortcuts should I know for editing?"
**Premiere Pro Guru:**
* Click the [**Premiere Pro Guru**](/conversation-starters/premiere-pro-guru) conversation starter
* Or type "help me with Premiere Pro" or "Premiere Pro question."
* Get specialized assistance for Premiere Pro workflows, troubleshooting, and techniques
* Access curated resources and tutorials specific to your question
***
### 2. Control Everything With Your Voice
**The Flow Killer:** Stopping to type commands or search for buttons interrupts your creative flow.
**The Flow Solution:** Talk to Chat Video Pro like a personal assistant—voice control keeps your hands on your keyboard and mouse.
Voice input transforms Chat Video Pro into a hands-free assistant:
* **Click the** [**microphone button**](/getting-started/interface-overview/voice-input) - In the composer, click the mic icon to enable voice input
* **Speak naturally** - Talk to Chat Video Pro as if speaking to a colleague
* **Get instant responses** - Your voice is transcribed and processed immediately
* **Stay in flow** - No need to stop typing or take your hands off your keyboard
**Workflow:**
1. While editing, you need to generate something or ask a question
2. Click the microphone button in the composer
3. Speak your request: "Generate a thumbnail for this video" or "How do I nest this sequence?"
4. Chat Video Pro processes your voice and responds
5. Continue working without breaking your editing rhythm
**Voice Use Cases:**
* "Generate an image of a coffee cup with a transparent background."
* "Create a thumbnail for this productivity video."
* "How do I apply a LUT in Premiere Pro?"
* "Remove the background from this image."
* "Help me write a prompt for video generation"
**Benefits:**
* **Faster than typing** - Speaking is quicker than typing for many people
* **Natural interaction** - Feels like talking to a personal assistant
* **Hands-free** - Keep your hands on the keyboard/mouse for editing
* **Maintains flow** - No context switching to type commands
***
### 3. Remove Backgrounds Instantly
**The Flow Killer:** Leaving Premiere Pro to use Canva, Photoshop, or sketchy online tools for background removal.
**The Flow Solution:** [Remove backgrounds](/features/image-generation/background-removal) from any image instantly, right inside Premiere Pro.
Chat Video Pro makes background removal as simple as asking:
* **Upload any image** - Drag and drop or upload the image you need
* **Say "Remove the background"** - That's it—no complex tools or subscriptions needed
* **Get true transparency** - Creates real transparent PNGs, not fake PNGs with white backgrounds
* **Super fast** - Background removal happens in seconds, not minutes
**Workflow:**
1. You have an image that needs a transparent background
2. Upload the image to Chat Video Pro (drag & drop or click upload)
3. Type or say: "Remove the background"
4. Get a true transparent PNG instantly
5. Download or save to Library for use in your project
**Use Cases:**
* **Downloaded graphics** - Remove backgrounds from images found online
* **Product shots** - Isolate products for compositing
* **Icons and logos** - Create transparent versions of graphics
* **Character isolation** - Remove backgrounds from photos for use in videos
* **Quick fixes** - Fix "fake PNGs" that have white backgrounds instead of transparency
**Benefits:**
* **No subscriptions** - No need for Canva Pro or Photoshop subscriptions
* **No sketchy websites** - Everything happens securely in your Premiere Pro extension
* **True transparency** - Real alpha channel transparency, not white backgrounds
* **Instant results** - Background removal in seconds
* **Stay in flow** - No need to leave Premiere Pro
***
### 4. Generate Graphics and Animate Them
**The Flow Killer:** Switching to Photoshop, Illustrator, or design tools to create graphics, then switching again to animate them.
**The Flow Solution:** Generate graphics with transparent backgrounds and animate them—all inside Premiere Pro.
Chat Video Pro lets you create and animate graphics in one continuous workflow:
* **Generate graphics with AI** - Create icons, title cards, and layered visuals instantly
* [**Create icons, title cards, and graphics** ](/features/image-generation/text-to-image)- Generate any visual element you need
* [**Animate with image-to-video**](/features/video-generation/image-to-video) - Use image-to-video models to bring your graphics to life
* [**Use Video Prompter Assistant**](/conversation-starters/video-prompter-assistant) - Get help writing optimized prompts to save time
**Workflow:**
**Step 1: Generate Your Graphic**
1. Enable Generate Media mode
2. Select an image model (Flux 2 Max or Nano Banana Pro for high-quality graphics; GPT Image 2 for canvas editor workflows)
3. Enter prompt: "Generate \[your graphic] on a clean white background."
* Example: "Generate an icon of a coffee cup with a transparent background"
* Example: "Create a title card that says 'Coming Soon' witha transparent background."
4. Generate your graphic
**Step 2: Animate Your Graphic (Optional)**
1. The generated graphic appears in your Library
2. Drag it into the composer
3. Switch to Video generation mode
4. Select an image-to-video model (Sora, Veo, Kling, etc.)
5. Use Video Prompter Assistant to help write your animation prompt
6. Generate an animated version of your graphic
**Pro Tip: Use Video Prompter Assistant**
* Click the **Video Prompter Assistant** conversation starter
* Describe what you want: "Animate a coffee cup icon with a subtle rotation."
* Get an optimized prompt with proven techniques
* Use that prompt for better animation results
**Use Cases:**
* **Icons** - Generate icons with transparent backgrounds for use in videos
* **Title cards** - Create animated title cards and lower thirds
* **Graphics additions** - Generate elements to add to existing graphics
* **Animated logos** - Create animated versions of logos or graphics
* **Visual elements** - Generate any graphic element you need for your edit
**Benefits:**
* **No design software needed** - Generate graphics without Photoshop or Illustrator
* **True transparency** - Real transparent backgrounds, not fake PNGs
* **Animate easily** - Turn static graphics into animated elements
* **Optimized prompts** - Video Prompter Assistant helps you write better prompts
* **Complete workflow** - Generate and animate without leaving Premiere Pro
***
### 5. Create Everything You Need for Publishing
**The Flow Killer:** Finishing your edit, then switching to multiple tools to create thumbnails, write descriptions, generate tags, and prepare publishing materials.
**The Flow Solution:** Complete your entire publishing workflow inside Premiere Pro—from video to published content.
Chat Video Pro transforms you from a video creator into a video strategist by enabling the complete publishing workflow:
* [**Brand Voice Assistant**](/conversation-starters/brand-voice-assistant) - Create SEO-optimized copy for your videos
* [**Thumbnail Mode** ](/features/image-generation/thumbnail-mode)- Generate optimized thumbnails with A/B testing
* **Complete publishing package** - Everything you need to send to clients or publish
**Workflow:**
**Step 1: Create SEO-Optimized Copy**
1. Click the [**Brand Voice Assistant**](/conversation-starters/brand-voice-assistant) conversation starter
2. Set up your brand profile (one-time setup, then reuse)
3. Upload your transcript or describe your video
4. Get optimized copy:
* YouTube titles (5 options, SEO-optimized)
* Full description (with keywords and structure)
* Chapters with timestamps
* Comma-separated tags
* Instagram/TikTok captions (if needed)
**Step 2: Generate Thumbnail**
1. Use [Frame Capture](/getting-started/interface-overview/frame-capture-button) to grab a key frame (optional)
2. Enable [Generate Media mode](/getting-started/interface-overview/generate-media-button)
3. Enable [Thumbnail Mode](/features/image-generation/thumbnail-mode) "On with Blueprints"
4. Describe your thumbnail concept
5. Generate 2-4 variations for A/B testing
6. Choose the best performing option
**Step 3: Complete Publishing Package**
* Video (your edit)
* Thumbnail (AI-generated, optimized)
* Title (SEO-optimized, multiple options)
* Description (keyword-rich, structured)
* Tags (comma-separated, ready to paste)
* Captions (for social media if needed)
**The Complete Value Proposition:** When you deliver this complete package to clients, you're not just a video creator—you're a video strategist. You're providing:
* Professional video content
* Optimized thumbnails that drive clicks
* SEO-optimized copy that helps discoverability
* Complete publishing materials ready to use
This is how you thrive in 2026: by providing complete value, not just video files.
**Benefits:**
* **Complete workflow** - Everything in one place, no context switching
* **Professional output** - SEO-optimized, platform-optimized, tested
* **Higher value** - You're a strategist, not just an editor
* **Client-ready** - Deliver complete packages, not just videos
* **Stay in flow** - Complete entire publishing workflow without leaving Premiere Pro
***
### Tips for Maintaining Flow State
#### Minimize Context Switching
* **Use voice control** - Speak instead of type to keep your hands on keyboard/mouse
* **Ask questions in-context** - Get help without leaving your project
* **Generate assets on-demand** - Create what you need when you need it
* **Complete workflows** - Finish entire publishing packages in one session
#### Leverage Assistants
* **Premiere Pro Guru** - For Premiere Pro questions and workflows
* **Video Prompter Assistant** - For optimized video generation prompts
* **Brand Voice Assistant** - For consistent, SEO-optimized copy
* **Story Cutter Assistant** - For transcript analysis and paper cuts
#### Build Your Library
* **Save successful assets** - Keep thumbnails, graphics, and elements in your Library
* **Create Elements** - Save character/product references for consistency
* **Reuse optimized prompts** - Learn what works and reuse it
* **Build templates** - Create reusable workflows for common tasks
#### Work in Batches
* **Generate multiple variations** - Create 2-4 thumbnails at once for A/B testing
* **Complete publishing packages** - Do all publishing tasks in one session
* **Batch questions** - Ask multiple Premiere Pro questions in one conversation
* **Plan ahead** - Identify what you'll need and generate it all at once
***
### The Complete Flow State Workflow
Here's how a complete editing session might look:
1. **Start editing** - Begin your Premiere Pro project
2. **Ask questions as needed** - Use Premiere Pro Guru for help without leaving
3. **Generate graphics on-demand** - Create icons, title cards, or elements as needed
4. **Remove backgrounds instantly** - Fix any graphics that need transparency
5. **Use voice control** - Speak commands to maintain flow
6. **Complete publishing package** - Generate thumbnail, copy, and tags when done
7. **Deliver a complete package** - Send everything to the client or publish
All without leaving Premiere Pro or breaking your creative flow.
***
### Next Steps
* [**Learn about Premiere Pro Guru**](/conversation-starters/premiere-pro-guru): Premiere Pro Guru Guide
* [**Understand Voice Input**](/getting-started/interface-overview/voice-input): Voice Input Feature
* [**See background removal**](/features/image-generation/background-removal): Background Removal Feature
* [**Learn about Brand Voice**](/conversation-starters/brand-voice-assistant): Brand Voice Assistant
* [**Understand Thumbnail Mode**](/features/image-generation/thumbnail-mode): Thumbnail Mode Feature
* [**Check pricing**](/getting-started/interface-overview/usage-panel): Pricing Information
***
**Related Workflows:**
* [**Generate AI B-Roll Directly Inside Premiere Pro**](/workflows/how-to-generate-b-roll-inside-premiere-pro-with-ai)
* [**AI VFX Inside Premiere Pro**](/workflows/how-to-create-ai-powered-visual-effects-in-premiere-pro)
* [**Story Editing with AI**](/workflows/how-to-cut-videos-faster-with-ai-assisted-story-editing-in-premiere-pro)
* [**AI Thumbnails in Premiere Pro**](/workflows/how-to-generate-high-ctr-thumbnails-inside-premiere-pro-with-ai)
# Project Folder Template
This is the same folder structure used internally by a six-figure video production company to manage long-term clients and large project libraries.
### Tutorial
{% embed url="" %}
The **Chat Video Pro Project Folder Template** is a professional-grade file and bin organization system designed to **save hours of setup time** and keep every video project clean, consistent, and easy to revisit — even years later.
***
### Why This Template Matters
Most editing slowdowns don’t come from editing — they come from:
* Searching for footage
* Rebuilding bins for every project
* Recreating sequences
* Re-organizing assets mid-edit
This template eliminates that friction by giving you:
* A consistent folder structure across every project
* Matching bins inside Premiere Pro and After Effects
* Pre-built sequences and compositions for all major formats
Once you use it, every project starts organized by default.
***
### What’s Included
When you download the Project Folder Template, you’ll receive:
* A **master project folder**
* A **Premiere Pro project file** with matching bins
* An **After Effects project file** with matching folders and comps
* Pre-made sequences and compositions for common resolutions and aspect ratios
Everything is designed to mirror itself:\
**Folder structure → Premiere bins → After Effects folders**
***
### How to Use the Template for a New Project
#### Step 1: Duplicate the Template
Whenever you start a new project:
1. Copy the entire template folder
2. Paste it into your working directory
3. Rename the folder to match your new project
***
#### Step 2: Rename the Project Files
Inside the **Projects** folder:
* Rename the Premiere Pro project file
* Rename the After Effects project file
This keeps everything aligned with the project name.
***
#### Step 3: Open the Premiere Pro Project
When you open the Premiere Pro file, you’ll see:
* Bins that **mirror the exact folder structure** on your drive
> Note: The template does **not automatically sync files**. You still import footage manually — but everything is already organized for you.
***
### Premiere Pro Bin Structure & Sequences
#### Matching Bins
Every major folder outside the project already exists as a bin inside Premiere Pro, so:
* Imports are clean
* Assets go exactly where they belong
* No time is wasted creating bins
***
#### Pre-Made Sequences
Inside the **Sequences** section, you’ll find:
* Landscape sequences
* Vertical (social) sequences
* Multiple resolutions (HD, 4K, etc.)
#### How to Use a Sequence Preset
1. Copy the sequence that matches your desired format
2. Paste it into your **Edits** folder
3. Rename it (e.g. `Social Vertical 01`)
4. Start editing immediately
No setup. No guesswork.
***
### After Effects Project Structure
The After Effects project includes:
* Matching folder organization
* A **Comp** section (similar to Premiere sequences)
* Pre-made compositions for:
* Vertical
* Landscape
* Multiple resolutions
This is especially useful for:
* Motion graphics projects
* Long-term brand work
* Large After Effects timelines
***
### Best Practices
* Always duplicate the template — never edit the original
* Keep footage imports consistent with the folder structure
* Use the provided sequences instead of creating new ones
* Store transcripts and exports in their designated folders
This consistency compounds over time and massively speeds up editing.
***
### Who This Is For
This template is ideal if you:
* Work with long-term clients
* Produce recurring content
* Manage multiple projects at once
* Want to stay organized without thinking about it
It’s especially powerful when combined with **Chat Video Pro Story tools**, since transcripts and exports stay easy to locate across projects.
***
### Common Questions
#### Does this automatically sync folders and bins?
No. You still import assets manually — the value is that **everything is already structured and ready**.
***
#### Can I customize the template?
Yes. You can modify it to fit your workflow, but we recommend keeping the core structure consistent.
***
#### Can I reuse this across all clients?
Absolutely. That’s the entire point — one system, every project.
***
### Summary
The Project Folder Template:
* Saves hours on every project
* Keeps your work organized long-term
* Removes friction from starting edits
* Helps you work faster and more professionally
This is foundational infrastructure for serious editors.
# Custom Export Presets
The Chat Video Pro Export Presets are custom Adobe Premiere Pro presets designed to prevent low-quality exports
### Tutorial
{% embed url="" %}
These presets are optimized to:
* Preserve detail after platform compression
* Reduce guesswork when exporting
* Save time by eliminating manual export setup
***
### Why These Presets Matter
Most export issues come from:
* Incorrect bitrate settings
* Platform compression is destroying quality
* Rebuilding export settings for every project
These presets give you **battle-tested settings** tuned specifically for:
* Social media platforms (Instagram, TikTok, Reels, Shorts)
* YouTube (with quality-optimized options)
* High-quality archival masters
* Transparent exports with alpha channels
Once installed, exporting becomes a one-click decision.
***
### How to Install the Export Presets
#### Step 1: Open Your File Browser
* macOS: Finder
* Windows: File Explorer
***
#### Step 2: Navigate to the Presets Folder
Go to:
```
Documents
→ Adobe
→ Adobe Media Encoder
→ [LATEST VERSION]
→ Presets
```
> This path is the same on both Windows and macOS.
***
#### Step 3: Copy the Presets
* Copy **all `.epr` files** from the download
* Paste them into the **Presets** folder
***
#### Step 4: Restart Premiere Pro
* Open Premiere Pro
* Go to the **Export** tab
* Scroll down to **More Presets**
* Under **Custom Presets**, you’ll see the CVP presets
You’re ready to export ✅
***
### Where to Find the Presets in Premiere Pro
1. Open the **Export** tab
2. Scroll to **More Presets**
3. Open **Custom Presets**
4. Select the appropriate **CVP\_** preset
***
### Preset Categories & When to Use Them
#### 🟢 CVP\_MASTER — High-Quality Archive Exports
Use these when:
* You’ve finished a project
* You want a clean, lossless version
* You may re-edit the project in the future
These exports are **large file sizes** but preserve maximum quality.
**Presets:**
* `CVP_MASTER_FULL HD 16:9` — 1920 × 1080
* `CVP_MASTER_4K UHD 16:9` — 3840 × 2160
* `CVP_MASTER_4K DCI` — 4096 × 2160
**Best practice:**\
Store these in your **Master** folder from the Project Folder Template.
***
#### 🟣 CVP\_SOCIAL — Social Media Exports
Use these for:
* Instagram
* TikTok
* Reels
* Shorts
These presets are tuned to:
* Hold up against platform compression
* Maintain sharpness and color
* Avoid over-compressed uploads
**Presets:**
* `CVP_SOCIAL_HORIZONTAL 4K 16:9` — 3840 × 2160
* `CVP_SOCIAL_HORIZONTAL FULL HD 16:9` — 1920 × 1080
* `CVP_SOCIAL_VERTICAL 4K 9:16` — 2160 × 3840
* `CVP_SOCIAL_VERTICAL FULL HD 9:16` — 1080 × 1920
***
#### 🟡 CVP\_YOUTUBE — YouTube & Web Uploads
Designed specifically for YouTube’s compression pipeline.
Each resolution includes:
* **Faster Export** → quicker turnaround
* **Slower Export** → higher fidelity (recommended when time allows)
The slower presets effectively process the video more thoroughly to preserve detail.
**Presets:**
* `CVP_YOUTUBE_4K UHD – Faster Export` — 3840 × 2160
* `CVP_YOUTUBE_4K UHD – Slower Export` — 3840 × 2160
* `CVP_YOUTUBE_FULL HD – Faster Export` — 1920 × 1080
* `CVP_YOUTUBE_FULL HD – Slower Export` — 1920 × 1080
**Recommendation:**\
If possible, upload in **4K** — even if your timeline is 1080p — for better compression results on YouTube.
***
#### ⚪ CVP\_TRANSPARENT — Alpha Channel Exports
Use this preset when you need:
* Transparency
* Overlays
* Rotoscoped elements
* Motion graphics exports
**Preset:**
* `CVP_TRANSPARENT (ProRes 4444+)`
This exports with an **alpha channel** and is ideal for compositing workflows.
***
### If You’re Not Sure Which Preset to Use
A quick rule of thumb:
* **Final archive?** → CVP\_MASTER
* **Instagram / TikTok?** → CVP\_SOCIAL
* **YouTube?** → CVP\_YOUTUBE (Slower if possible)
* **Transparency needed?** → CVP\_TRANSPARENT
You can also refer to the **PDF guide** included with the download for a quick reference.
***
### Best Practices
* Match the preset to your **sequence resolution**
* Don’t stack unnecessary export effects
* Keep color management consistent
* Test one export before batch exporting
Used correctly, these presets eliminate most quality complaints.
***
### Summary
The Chat Video Pro Export Presets:
* Prevent low-quality uploads
* Remove export guesswork
* Improve platform compression results
* Save time on every delivery
If you care about how your work looks online, these presets are essential.
***
**Next:** Learn how these presets integrate with the [**Project Folder Template**](/workflows/project-folder-template) for a clean end-to-end workflow.
####
# Usage Tips & Shortcuts
This page highlights small but powerful tips that can significantly improve your day-to-day experience with Chat Video Pro.
### 🔹 Use the Tilde (\`) Key for Full Screen View
If you ever feel like:
* The panel feels cramped
* A generation looks cut off
* You want a clearer view of images or video previews
You can instantly **maximize the Chat Video Pro panel**.
#### How it works
* Press the **tilde (\`)** key while your cursor is over the Chat Video Pro panel
* The panel will expand to full screen
This is especially useful when reviewing:
* Generated images
* Video previews
* Longer responses or instructions
#### If the tilde key doesn’t work
Some keyboard layouts or system settings may block the tilde shortcut.
In that case:
* **Double-click the panel title that says “Chat Video Pro”**
* The panel will expand to full screen manually
***
### 🔹 Don’t Know How to Do Something? Ask Chat Video Pro
Chat Video Pro is **fully trained on its own capabilities**.
If you’re unsure:
* How to perform a task
* Which workflow to use
* Which video or image model to select
Just **ask Chat Video Pro directly**.
#### Examples
* “How do I generate a social media transition?”
* “Which video model should I use for this clip?”
* “What’s the best way to remove a background?”
It can:
* Explain how to do things inside the tool
* Suggest which models are appropriate
* Help guide you through workflows step by step
Think of it as built-in documentation you can talk to.
***
### 🔹 Drag & Drop Content Back Into the Composer
You can **reuse content directly inside Chat Video Pro** without re-importing files manually.
#### What you can drag and drop
* Images generated in chat
* Video clips from the chat
* Assets from the Chat Video Pro library
#### Why this is useful
* Iterating on a generation
* Using an image as input for image-to-video
* Refining or expanding an existing result
Simply drag the asset back into the composer area and continue working from it.
***
### 🔹 Why These Small Tips Matter
These shortcuts help you:
* Work faster
* See results more clearly
* Reduce friction while experimenting
* Stay inside one workflow instead of switching tools
As Chat Video Pro grows, this page will continue to evolve with more quality-of-life improvements and power-user tips.
# Pricing FAQ
This page answers common questions about Chat Video Pro pricing, generation costs, and billing, and is designed to clear up confusion around how pay-as-you-go AI works.
#### **Is Chat Video Pro a subscription?**
No. Chat Video Pro requires a **one-time license purchase**.
Once purchased, you can use the software indefinitely on the licensed machine(s). There are **no monthly fees** for Chat Video Pro itself.
***
#### **Why do I need to pay for generations after buying the license?**
Chat Video Pro does not host or charge for AI models directly.
All AI operations (video generation, image generation, VFX, upscaling, etc.) run through **fal.ai**, which uses a **pay-as-you-go** billing model.
This means:
* You only pay when you generate
* There are no usage caps
* There is no markup from Chat Video Pro
* You are billed directly by the provider
***
#### **What am I actually paying fal.ai for?**
You are paying for:
* GPU compute
* Model inference
* Video/image processing time
Chat Video Pro simply acts as a **bridge inside Premiere Pro** that sends requests to the model provider using your credentials.
***
#### **How much do generations cost?**
Costs vary by:
* Model
* Resolution
* Duration
* Operation type (image, video, tracking, upscaling, etc.)
Typical ranges:
* Images: a few cents
* Short video clips: cents to a few dollars
* Longer or higher-resolution operations: more
> **Important:** Prices can change over time. Always refer to the [Fal.ai pricing page](https://fal.ai/pricing) for current rates.
***
#### **Where can I see exactly what I’m being charged?**
You have two places to monitor usage:
**Inside Chat Video Pro**
* Gear icon → Usage panel
**Inside your fal.ai account**
* [fal.ai/dashboard/usage-billing](https://fal.ai/dashboard/usage-billing)
The provider dashboard is the source of truth for billing.
***
#### **Do unused credits expire?**
No. Credits added to your provider account remain until used.
***
#### **Can I set spending limits or alerts?**
Yes. Inside your provider account, you can:
* Monitor usage in real time
* Set budget alerts
* Control how much you load into your balance
This gives you full control over spending.
***
#### **How can I avoid wasting credits?**
Best practices:
* Use preview or tracking modes first when available (for example, SAM 3 “Track Frame”)
* Start with shorter clips
* Keep prompts clear and specific
* Test with lower-cost models before scaling up
Detailed cost-saving tips are covered in the **Usage & Best Practices** section.
***
#### **Can I use Chat Video Pro commercially?**
Yes. You retain full rights to anything you generate using Chat Video Pro and connected models.
***
#### **What is refundable?**
We offer a **seven-day money-back guarantee** for the **base Chat Video Pro package**.
Digital bonus assets (such as templates or presets) may be non-refundable.
If something isn’t working as expected, we strongly recommend contacting support before requesting a refund — most issues are configuration-related and can be resolved quickly.
***
#### **Who do I contact for billing or pricing questions?**
* In-app: Gear icon → Contact Us
* Email: Support contact listed in the About panel
***
**Next:** Learn how to monitor and manage usage in the **Usage Dashboard**.
{% content-ref url="/pages/8teMzAnXOweDGrYD9JBK" %}
[Usage Panel](/getting-started/interface-overview/usage-panel)
{% endcontent-ref %}
**CVP works best when it's built around how you edit — not the other way around.** A Founder Onboarding Session is 60 minutes live inside your Premiere timeline, configured for your real projects. [BOOK YOUR SESSION →](https://checkout.chatvideopro.com/checkout/buy/6b983504-8882-45e4-9ca8-64ea1a28eddf)
# Installation & Setup FAQ
This page answers common questions and issues related to installing, activating, and setting up Chat Video Pro inside Adobe Premiere Pro. If something isn’t working right after install, start here.
### **What do I need before installing Chat Video Pro?**
Before installing, make sure you have:
* A supported version of **Adobe Premiere Pro**
* A stable internet connection
* A valid Chat Video Pro license key
* A **fal.ai** account with credits
Chat Video Pro runs inside Premiere Pro and requires access to external AI models to function fully.
***
### **Does Chat Video Pro work on both macOS and Windows?**
Yes. Chat Video Pro supports **both macOS and Windows**.
If you experience platform-specific issues:
* Restart your system
* Reinstall the extension
* Confirm OS permissions are not blocking extensions
***
### **Where do I enter my license key?**
Your license key is entered during the setup process inside Chat Video Pro.
License keys are delivered via email after purchase.\
If you don’t see the email, check your spam folder.
***
### **Why do I need a fal.ai account?**
Chat Video Pro does not process AI generations itself.
All image, video, and AI operations are handled by **fal.ai**, which provides:
* Pay-as-you-go billing
* Access to advanced AI models
* Transparent usage tracking
Without a connected fal.ai account, generations will fail.
***
### **How do I connect my fal.ai API key?**
1. Create an account at fal.ai
2. Add credits to your account
3. Copy your API key (`key_id:key_secret`)
4. Paste it into Chat Video Pro during setup or change it in the settings menu
Once saved, restart Premiere Pro to ensure the connection is active.
***
### **I installed Chat Video Pro, but the panel doesn’t appear**
Try the following:
1. Restart Premiere Pro
2. Open **Window → Extensions / Extensions (Legacy)** (depending on version)
3. Reinstall the extension
4. Restart your computer
If the panel still doesn’t load, confirm your Premiere Pro version is supported.
{% content-ref url="/pages/DlgpyymQR1sTGaAeyHk4" %}
[Compatibility](/getting-started/compatibility)
{% endcontent-ref %}
***
### **The panel opens, but nothing happens when I click Generate**
Common causes:
* Missing or invalid API key
* No credits in your fal.ai account
* Unsupported model selected
* Premiere Pro temporarily frozen
**Fix:**
* Confirm credits are available
* Re-paste your API key
* Restart Premiere Pro
* Try a simpler prompt first
***
### **I’m getting a “403” or “Unauthorized” error**
This usually means:
* Your fal.ai account has no funds
* The API key is incorrect
* Permissions were revoked or expired
**Fix:**
1. Log into your fal.ai dashboard
2. Add credits
3. Re-enter your API key
4. Restart Premiere Pro
***
### **Can I install Chat Video Pro on multiple computers?**
Each device requires its own seat.
If you need to use Chat Video Pro on multiple machines, purchase additional seats.
***
### **Do I need to be logged into the internet at all times?**
Some features work locally, but **AI generation requires an active internet connection**.
If you go offline:
* Local tools may still function
* Generations will fail until connectivity is restored
***
### **Is antivirus or system security blocking the extension?**
In some cases, system security tools may block extensions from running correctly.
If you experience unexplained behavior:
* Temporarily disable aggressive antivirus tools
* Whitelist the Chat Video Pro extension directory
* Reinstall and restart
***
### **I’m stuck during setup — what should I do?**
If setup doesn’t complete:
* Restart Premiere Pro
* Double-check license key and API key
* Confirm credits are available
* Try reinstalling the extension
Most setup issues are resolved with a clean restart and valid credentials.
***
### **How do I contact support if installation fails?**
* In-app: Gear icon → Contact Us
* Email: Support at
When contacting support, include:
* OS (macOS or Windows)
* Premiere Pro version
* Error message (if any)
* What step you’re stuck on
***
**Next:** Learn how to monitor usage and billing in the **Usage Dashboard**.
{% content-ref url="/pages/8teMzAnXOweDGrYD9JBK" %}
[Usage Panel](/getting-started/interface-overview/usage-panel)
{% endcontent-ref %}
**Prefer to have it built with you?** If you'd rather skip the guesswork, book a Founder Onboarding Session. We go through setup and your first workflow live inside Premiere — 60 minutes, your project, your pipeline. [BOOK YOUR SESSION →](https://checkout.chatvideopro.com/checkout/buy/6b983504-8882-45e4-9ca8-64ea1a28eddf)
# Generation Errors & Failed Jobs FAQ
This page covers common issues related to failed generations, missing results, errors, and unexpected behavior when generating images or videos with Chat Video Pro.
Most failed generations come from one of five causes:
* The provider account has no credits, no valid API key, or no permission.
* The selected model needs a different input type.
* The source image or video does not meet the workflow limits.
* The provider finished the job, but the result did not return to the panel.
* The provider had a temporary outage, timeout, or queue issue.
Start with the checks below before assuming the result is lost.
***
#### What should I check first?
Use this quick checklist:
1. Confirm your API key is saved.
2. Confirm your provider account has credits.
3. Check whether the job appears in [fal.ai recent history](https://fal.ai/dashboard/recent-history).
4. Make sure the workflow has the required inputs.
5. Check the source media length, format, and aspect ratio.
6. Try a shorter clip or simpler prompt.
7. Restart Premiere Pro if the panel is stuck.
If the job appears in your provider dashboard, the provider received it. If the result did not return to Chat Video Pro, support can usually diagnose it from the job time, model, and error message.
***
#### Credits were spent, but I did not get a result
This usually means the generation ran, but the result did not make it back into the panel.
Common causes:
* Premiere Pro became unresponsive.
* The Chat Video Pro panel refreshed.
* The network connection dropped after the provider accepted the job.
* The provider completed the job after the local request timed out.
* The result file was created, but the panel did not receive the final URL.
Check [fal.ai recent history](https://fal.ai/dashboard/recent-history). If the job appears there, open it and confirm whether an output exists.
If an output exists, the generation was not lost. If the provider charged credits but no output exists, contact support with the job details.
***
#### Nothing happens when I click Generate
Try these in order:
1. Confirm your API key is saved.
2. Confirm your provider account has credits.
3. Confirm the selected model supports the input you attached.
4. Remove extra attachments that are not needed.
5. Try a shorter prompt.
6. Restart Premiere Pro.
If the request still does not start, check whether the model list looks correct. If no compatible models appear, the composer may be receiving the wrong attachment type or the workflow may require different media.
See Model Selection.
***
#### I see 403, Unauthorized, or Permission Denied
This is usually an account or API-key issue.
Common causes:
* Your provider account has no credits.
* The API key is missing, invalid, or copied incorrectly.
* The provider revoked or changed access permissions.
* The model requires access your account does not currently have.
Fix:
1. Log into your provider dashboard.
2. Add credits if needed.
3. Re-copy the full API key.
4. Paste it into Chat Video Pro settings.
5. Restart Premiere Pro.
6. Try a small test generation.
***
#### Error: At least one image URL is required
This means the selected model or workflow needs an image, but no usable image reached the generation request.
Common examples:
* You selected an Image-to-Video model without attaching an image.
* You opened Motion Director without loading a source image.
* You opened AI Transitions but only provided one frame.
* You selected an image editing model without a source image.
Fix options:
* Capture a frame from the timeline.
* Upload or import an image.
* Use a recent image from the asset loader.
* Switch to a text-only model if you are starting from a prompt.
See Image-to-Video, Motion Director, and AI Transitions.
***
#### Studio says my asset is not supported
Studio workflows filter assets on purpose. Each workflow only accepts the input types it can process.
Examples:
| Workflow | Required input |
| --------------- | -------------------------------------- |
| Cinematic Lab | Text prompt, optional reference images |
| Motion Director | Image |
| AI Transitions | Start image and end image |
| Rotoscope | Video |
| Erase Objects | Video |
| Add Effects | Video |
| Reshoot | Video |
| Upscale | Video |
| Motion Capture | Motion video plus character image |
| Multi-Cam | Image or video |
| Relight Scene | Image or supported video |
If an asset is hidden or disabled, it may be the wrong media type, too long, too large, or not available to that workflow.
See Studio.
***
#### My video is too long
Many AI video editing tools work best on short clips. Some have hard limits.
Important limits:
Workflow
Practical limit
Erase Objects
5 seconds maximum
Add Effects
At least 3 seconds; shorter focused clips work best
Reshoot
2-20 second selected segment
Generative Extend
Input video must be 23 seconds or shorter
Motion Capture
Use a short, clear motion reference
Rotoscope
Shorter clips are faster and easier to verify
Fix:
1. Trim the clip around the exact moment you need.
2. Avoid sending extra lead-in or tail frames.
3. Run the creative edit first.
4. Upscale only after the result is approved.
See Erase Objects, Reshoot, and Generative Extend.
***
#### The generation took a long time and then failed
Long jobs are more likely to fail than small tests.
Common causes:
* Long source video.
* High resolution.
* Complex prompt.
* Multiple references.
* Provider queue load.
* Network interruption.
Best practice:
* Test with a shorter clip first.
* Use a smaller draft setting when available.
* Use one main creative instruction per generation.
* Avoid combining cleanup, style transfer, VFX, and upscaling in one pass.
* Retry later if the provider dashboard shows widespread failures or stuck jobs.
***
#### My transition failed or looks wrong
Transition jobs are sensitive to the relationship between the two frames.
Check:
* Do both images have the same aspect ratio?
* Are the start and end frames visually compatible?
* Is the subject in a similar position?
* Is the transition style too complex for the duration?
* Did you accidentally attach reference images instead of a start/end pair?
Fix:
1. Crop or regenerate the frames to the same aspect ratio.
2. Keep the subject placement similar.
3. Use a simpler transition prompt.
4. Use Studio AI Transitions for a guided version.
See Transition Mode.
***
#### My image-to-video result changed the subject too much
Image-to-video models can drift if the source image is unclear or the prompt asks for too much change.
Fix:
* Use a clean, high-quality source image.
* Keep the main subject large enough in frame.
* Ask for motion, not redesign.
* Match the source aspect ratio when possible.
* Use Motion Director when the task is mainly camera movement.
Example:
Good: `Slow dolly push in, subject remains identical, background parallax, cinematic lighting.`
Risky: `Make this person dance in a new outfit on a different planet with a new hairstyle.`
***
#### My reference images are ignored
Reference images work best when each reference has a clear job.
Check:
* Did you choose a reference-capable model?
* Are the references clean and easy to read?
* Did you tell the prompt what each reference is for?
* Are you asking the model to preserve too many details at once?
Better prompt:
`Use Image 1 for the character identity, Image 2 for the jacket design, and Image 3 for the lighting style. Create a 5-second hero shot with the same character and jacket.`
See Reference Mode.
***
#### My result has the wrong aspect ratio or crop
Aspect ratio problems usually come from mismatched source media or changing formats mid-workflow.
Fix:
* Choose the final delivery aspect ratio before generating.
* Use source images that already match the intended output shape.
* Keep transition start/end frames the same shape.
* For vertical-first work, generate vertical sources instead of cropping a wide result later.
* If a model does not support the aspect ratio you need, choose another model or reframe in Premiere.
See Text-to-Video, Image-to-Video, and Text-to-Image.
***
#### My result appears once, then disappears
This can happen if Premiere Pro refreshes the extension panel or the session resets.
Check:
1. The current chat.
2. The Library.
3. [fal.ai recent history](https://fal.ai/dashboard/recent-history).
If the provider dashboard has the output, the job completed. If Chat Video Pro did not keep the result in the current session, download it from the provider dashboard or contact support with the job link.
***
#### The model I want is missing or disabled
The model list changes based on your inputs and settings.
Common reasons:
* A text-only model is hidden because an image or video is attached.
* An image-to-video model is hidden because no image is attached.
* Transition Mode appears because two images are attached.
* Reference models appear only when the selected model supports references.
* Some options are disabled when duration, resolution, aspect ratio, or media length is not supported.
Remove extra attachments or switch workflows if the model list does not match what you expected.
See Model Selection.
***
#### The generation worked before, but not now
Check:
* Do you still have credits?
* Did your API key change?
* Did the source media change?
* Did you attach a different number of images?
* Is the clip longer than before?
* Is the model temporarily unavailable from the provider?
If the same prompt and media worked earlier, try a small test generation. If the small test works, the issue is probably the current media, prompt, duration, or settings.
***
#### What should I send to support?
Include as much of this as possible:
* What you were trying to do.
* The workflow or model used.
* Approximate time of generation.
* Error message, if shown.
* Whether the job appears in [fal.ai recent history](https://fal.ai/dashboard/recent-history).
* Source media type: text, image, video, two images, references, or Studio workflow.
* Source video duration if video was used.
* Output settings: aspect ratio, duration, resolution, audio, or upscale factor.
* OS and Premiere Pro version.
* Screenshot of the error or stuck state.
This helps support identify whether the issue is local setup, provider account, input mismatch, model constraint, or a provider-side failure.
**Hitting a wall?** A Founder Onboarding Session cuts through it. We open your project live, fix whatever is blocking you, and build your AI workflow so you leave running, not troubleshooting. 60 minutes. [BOOK YOUR SESSION ->](https://checkout.chatvideopro.com/checkout/buy/6b983504-8882-45e4-9ca8-64ea1a28eddf)
# How Do I…(Quick Tasks)
This page answers common “How do I do X?” questions in Chat Video Pro. Each answer gives you the quick path, with links to deeper documentation where relevant.
#### How do I chat vs generate content?
The simple rule:
* Use **Chat** for questions, planning, Premiere help, and quick images.
* Use **Generate Media** for direct image and video model control.
* Use **Studio** for guided creative workflows like Cinematic Lab, Motion Director, AI Transitions, Avatar Studio, Rotoscope, Erase Objects, Add Effects, Reshoot, Upscale, Motion Capture, Multi-Cam, Relight Scene, and Reframe.
For the full routing guide, see Model Selection.
***
#### How do I open Studio?
1. Open Chat Video Pro inside Premiere Pro.
2. Click **Studio** in the sidebar.
3. Choose a workflow card from the Launchpad.
Use Studio when the task has a known shape: cinematic stills, camera moves, transitions, cleanup, relighting, effects, reshoots, upscaling, or motion transfer.
See Studio.
***
#### How do I know whether to use Studio or Generate Media?
Use **Generate Media** when you want to choose the model and settings directly.
Use **Studio** when you want Chat Video Pro to guide the workflow.
Examples:
| Goal | Best starting point |
| ----------------------------------------------- | ---------------------- |
| Generate a video from a prompt | Text-to-Video |
| Animate a still with a direct prompt | Image-to-Video |
| Animate a still with camera movement presets | Studio Motion Director |
| Create a transition between two frames manually | Transition Mode |
| Create a guided AI transition | Studio AI Transitions |
| Clean up or finish existing footage | Studio |
***
#### How do I generate a thumbnail?
Fast path:
1. Enable **Generate Media**.
2. Choose an image model.
3. Turn on Thumbnail Mode.
4. Prompt the topic, promise, subject, emotion, and composition.
5. Generate 2-4 directions.
For more polished thumbnails:
* Use Frame Capture to match a real moment from your edit.
* Use Studio Cinematic Lab for a premium hero frame.
* Use GPT Image 2 when text or layout precision matters.
* Use Canvas Editor for final layout edits.
* Use Image Upscaling after the final thumbnail is approved.
See High-CTR Thumbnail Workflow.
***
#### How do I create a cinematic still or key art?
Use Studio Cinematic Lab.
1. Open **Studio**.
2. Choose **Cinematic Lab**.
3. Describe the scene.
4. Choose camera, lens, focal length, aperture, aspect ratio, resolution/quality, and model.
5. Generate a batch.
6. Send the best image back to chat.
Use Cinematic Lab for thumbnails, key art, lookdev frames, hero images, story frames, and source images for later video workflows.
***
#### How do I animate a still image?
You have two good paths.
Use Image-to-Video when you want direct model control and a normal motion prompt.
Use Studio Motion Director when you want camera movement presets such as push in, dolly, orbit, crane, handheld, or drone moves.
Fast Studio path:
1. Open **Studio**.
2. Choose **Motion Director**.
3. Load a still image.
4. Choose a motion preset.
5. Generate.
***
#### How do I generate video from text?
1. Enable **Generate Media**.
2. Choose a text-to-video model.
3. Set duration, aspect ratio, resolution, and audio options.
4. Write a prompt with subject, setting, action, camera, style, and mood.
5. Generate.
See Text-to-Video and Supported Video Models.
***
#### How do I generate a transition?
Use Transition Mode when you want direct model control.
Use Studio AI Transitions when you want guided transition styles and better start/end-frame prompting.
Fast Studio path:
1. Open **Studio**.
2. Choose **AI Transitions**.
3. Load a start frame and end frame.
4. Choose a transition style.
5. Generate the transition.
***
#### How do I remove a background from an image?
Use Background Removal.
1. Attach or upload an image.
2. Ask: `Remove the background.`
3. If the image has multiple subjects, say what to keep: `Keep only the red sneaker and make everything else transparent.`
4. Save or reuse the transparent PNG.
Use this for still images only.
***
#### How do I remove a background from video?
Use Studio Rotoscope.
1. Open **Studio**.
2. Choose **Rotoscope**.
3. Load a video.
4. Select the subject with Text, Box, or Point.
5. Click **Track Frame** to preview the mask.
6. Click **Track Entire Video**.
7. Click **Remove Background**.
Background Removal is for images. Rotoscope is for video.
***
#### How do I erase an object from video?
Use Studio Erase Objects.
1. Open **Studio**.
2. Choose **Erase Objects**.
3. Load a short video.
4. Describe or point to the object that should disappear.
5. Generate the cleaned clip.
Important: Erase Objects uses **VOID**. Trim the clip to the problem area when you can for the best results.
***
#### How do I add VFX without After Effects?
Use Studio Add Effects.
1. Open **Studio**.
2. Choose **Add Effects**.
3. Load a video.
4. Prompt the effect: `Add heavy rain with wet streets` or `Add cinematic fog and blue moonlight.`
5. Optional: add reference images for style or subject guidance.
6. Generate and compare before/after.
Use Add Effects for rain, fire, fog, atmosphere, stylized looks, energy effects, and broad shot transformations.
***
#### How do I reshoot one part of a clip?
Use Studio Reshoot.
1. Open **Studio**.
2. Choose **Reshoot**.
3. Load a video.
4. Select the segment you want to change.
5. Describe what should happen differently.
6. Choose whether to replace video, audio, or both.
7. Generate.
Use Reshoot for targeted changes, not full-clip style transfer.
***
#### How do I upscale an image?
Use Image Upscaling.
1. Start with a generated or uploaded image.
2. Click **Transform**.
3. Choose an image upscaler.
4. Generate the higher-resolution image.
Upscale after the image is approved. Do not upscale every draft.
***
#### How do I upscale a video?
Use Studio Upscale.
1. Open **Studio**.
2. Choose **Upscale**.
3. Load a video.
4. Choose an upscale model and scale factor.
5. Generate the upscaled result.
Upscale video after creative edits are finished.
***
#### How do I relight an image or short video?
Use Studio Relight Scene.
1. Open **Studio**.
2. Choose **Relight Scene**.
3. Load an image or supported video.
4. Choose light direction and style.
5. Generate the relit result.
Use Relight Scene for changing light direction, mood, time of day, or atmosphere while preserving the scene.
***
#### How do I change a clip's aspect ratio?
Use Studio Reframe.
1. Open **Studio**.
2. Choose **Reframe**.
3. Load a video clip.
4. Select the target aspect ratio (for example, 9:16 for vertical social).
5. Generate the reframed result.
Reframe uses AI edge fill to extend or crop the frame intelligently. Use it when repurposing 16:9 footage as 9:16 for Reels, Shorts, or TikTok, or any time the output format requires a different aspect ratio than the source.
You can also access Reframe from the **Transform** menu that appears under a generated video clip in chat.
***
#### How do I generate alternate camera angles?
Use Studio Multi-Cam.
1. Open **Studio**.
2. Choose **Multi-Cam**.
3. Load an image or video.
4. Choose single angle or grid output.
5. Generate alternate views.
Use Multi-Cam when you need more coverage from one existing image or clip.
***
#### How do I transfer motion onto a character image?
Use Studio Motion Capture.
1. Open **Studio**.
2. Choose **Motion Capture**.
3. Load a motion reference video.
4. Load a character image.
5. Generate the motion transfer.
Use this when you have movement you like and a separate character image you want to animate.
***
#### How do I create a talking-head avatar video?
Use Studio Avatar Studio.
1. Open **Studio**.
2. Choose **Avatar Studio** (Production department).
3. Upload a photo of the presenter or choose one from Recents.
4. Provide a script or prompt for the spoken content.
5. Generate the talking-head video.
Use Avatar Studio for UGC-style ads, spokesperson clips, social content, and any project that needs a presenter without a camera crew.
***
#### How do I capture a frame from my timeline?
1. Move the Premiere playhead to the frame you want.
2. Click Frame Capture in Chat Video Pro.
3. The captured frame is added to the composer.
Use captured frames for:
* Cinematic Lab references.
* Thumbnail backgrounds.
* Image-to-Video source frames.
* AI Transitions start/end frames.
* Background removal.
* Reference images for visual consistency.
***
#### How do I use Story Cutter for long footage?
Use Story Cutter Assistant.
For best results:
1. Export your transcript from Premiere Pro's **Text** panel as a `.json` transcript file.
2. Attach the `.json` transcript in the Story Cutter conversation.
3. Include platform, target runtime, tone, and the goal of the edit.
Story Cutter is best for long interviews, podcasts, documentary footage, webinars, and narrative-heavy source material.
***
#### How do I avoid wasting credits?
Best practices:
* Start with shorter clips and smaller batches.
* Use still images to test visual direction before generating video.
* Use Cinematic Lab to design a frame before animating it.
* Use Track Frame before full Rotoscope processing.
* Upscale only after the creative result is approved.
* Generate 2-4 image options when exploring, then refine the strongest one.
* Use cheaper/faster models for early drafts and premium models for final passes.
See Pricing FAQ for billing basics.
***
#### How do I find something I generated?
Check three places:
1. The current chat result.
2. The Library.
3. Your provider dashboard: [fal.ai/dashboard/recent-history](https://fal.ai/dashboard/recent-history).
If a result did not return to the panel, it may still exist in your provider history.
***
#### How do I get help if I am stuck?
Start here:
1. Check Generation Errors & Failed Jobs FAQ.
2. Confirm your API key and credits are active.
3. Restart Premiere Pro.
4. Try a simpler prompt, shorter clip, or supported input type.
If you contact support, include:
* What you were trying to do.
* The workflow or model used.
* Source media type and duration.
* Error message, if any.
* OS and Premiere Pro version.
* Approximate time of the failed generation.
**Rather have your workflow built live?** Book a Founder Onboarding Session. We configure Chat Video Pro around your real project in 60 minutes, inside your Premiere timeline, with the founder, so you are running instead of searching. [BOOK YOUR SESSION ->](https://checkout.chatvideopro.com/checkout/buy/6b983504-8882-45e4-9ca8-64ea1a28eddf)
# Trouble Accessing Downloads
#### Make sure you are signed in with the email account you used when purchasing.
### ✅ Sign In
Follow these steps to access your files:
***
#### **Step 1: Open your order**
* Click the order link in your purchase email\
**OR**
* Go directly here:\
👉
***
#### **Step 2: Sign in to your account**
* If prompted, log in using the same email you used at checkout
* This step is important — your files won’t show correctly unless you’re signed in
***
#### **Step 3: Access your downloads**
* Once logged in, your correct files will appear
* You can now download everything normally
***
### Why this happens
You must be logged in with your account email to access your downloads.
***
### 💬 Still stuck?
If you’re still having trouble accessing your files, feel free to reach out, and we’ll help you get set up right away. Support email: