Video Generation
Generate final videos from your approved storyboard shots.
Before You Begin
Preparation Checklist
Ensure your storyboard is ready:
- All hero frames approved
- Character/subject consistency verified
- Shot sequence finalized
- Durations set correctly
- Sufficient credits available
Generating Videos
Single Shot Generation
Generate video for individual shots:
- Select a shot with status
image_ready - Click Approve & Generate Video in the Shot Details panel
- The shot status changes to
generating_video - Wait for generation to complete
- Review the result in the video preview tab
How Video Generation Works
For each shot, the system:
- Resolves the model -- Maps the configured model name to the full API path using the model resolver
- Validates duration -- Clamps the shot's target duration to the model's allowed values (nearest valid duration)
- Enhances the prompt -- Injects cinematography details (camera angle, motion, lens, framing, lighting, environment FX, motion intensity) into the base prompt
- Generates via API -- Sends the hero frame image + enhanced prompt to the video generation edge function
- Polls for completion -- Async polling until the video is ready
- Updates the shot -- Stores the video URL and updates status to
complete
Video Models
Standard Models (Default)
| Model | Key | Audio | Duration | Best For |
|---|---|---|---|---|
| Veo 3 | veo | Yes | 4, 6, 8s | Highest quality, photorealism, cinematic |
| Veo 3 Fast | veo3-fast | Yes | 4, 6, 8s | Faster Veo 3 generation |
| Sora | sora | No | Flexible | Cinematic, complex motion, narrative coherence |
| Waver 1.0 | waver | No | 5-10s | Smooth motion, natural movement, general purpose |
| Hailuo 2.3 | hailuo | No | 6, 10s | Fast generation, quick iterations |
| Hailuo Pro | hailuo-pro | No | 6, 10s | Higher quality Hailuo |
| Hailuo Fast | hailuo-fast | No | 6, 10s | Fastest Hailuo tier |
| Wan 2.5 | wan | No | 3-10s | Flexible duration range |
| Wan Fast | wan-fast | No | 3-10s | Faster Wan generation |
| Kling O1 | kling | No | 5, 10s | Camera control, character consistency |
| Kling 2.6 Pro | kling-2.6 | Yes | 5, 10s | Native audio, 50% cheaper than Veo 3 |
| Seedance | seedance | No | 5-10s | Alternative generation |
Budget Models
When Budget Mode is enabled, auto resolves to cost-effective alternatives:
| Type | Standard Auto | Budget Auto |
|---|---|---|
| Image | Nano Banana Pro | SeedDream v4.5 |
| Video | Veo 3 | Kling 2.6 Pro |
Audio Support
Models with native audio generate synchronized sound effects and ambient audio alongside the video:
| Model | Audio |
|---|---|
| Veo 3 / Veo 3 Fast | Yes |
| Kling 2.6 Pro | Yes |
| All others | No |
When Require Sound is enabled in project settings and the model is set to auto, the resolver prefers audio-capable models.
Duration Handling
Each model has specific duration constraints. When a shot's target duration doesn't match the model's allowed values, the system automatically selects the closest valid duration:
| Model | Allowed Durations |
|---|---|
| Veo 3 / Veo 3 Fast | 4, 6, 8 seconds |
| Waver 1.0 | 5, 6, 7, 8, 9, 10 seconds |
| Hailuo (all tiers) | 6, 10 seconds |
| Wan 2.5 (both speeds) | 3, 4, 5, 6, 7, 8, 9, 10 seconds |
| Kling O1 | 5, 10 seconds |
| Kling 2.6 Pro | 5, 10 seconds |
| Seedance | 5, 6, 7, 8, 9, 10 seconds |
Example: A shot targeting 7 seconds on Hailuo will be generated as 6 seconds (nearest valid duration).
Prompt Enhancement
The video prompt is automatically enhanced with cinematography details from the storyboard pipeline:
Base prompt: "Athletic woman running through empty city street at dawn"
Enhanced: "Athletic woman running through empty city street at dawn.
Camera: low 3/4 hero angle, smooth dolly following subject.
Lens: 85mm portrait compression.
Framing: hero product center-frame.
Lighting: golden-hour backlight with rim highlights.
Motion intensity: medium."
This ensures the video inherits the full cinematography direction designed by the pipeline workers.
Extending Videos
Extend a completed video beyond its original duration:
- Select a shot with a completed video
- Click Extend Video in the Shot Details panel
- Configure:
- Model: Choose the video model for the extension
- Resolution: Select output resolution
- Prompt: Optional prompt for the extension (auto-generated if empty)
- The extension chains onto the existing video
Extensions are tracked as a chain on the shot. Each extension records:
- Parent video URL (source that was extended)
- Result video URL
- Model and resolution used
- Processing status and job ID
You can remove individual extensions from the chain if needed.
Reviewing Generated Videos
Playback
Click any video preview to open the Media Lightbox for full-size playback. In the Shot Details panel, when both image and video are available, tabs let you switch between the Image and Video views.
Quality Check
Review each video for:
| Aspect | What to Check |
|---|---|
| Motion | Smooth, natural movement matching cinematography |
| Subject | Consistent with hero frame appearance |
| Artifacts | Glitches, distortions, or unnatural elements |
| Audio | Sync and appropriate sounds (for audio-capable models) |
| Duration | Correct length for the shot |
Version History
Each regeneration creates a version. You can:
- Browse previous versions in the thumbnail carousel
- Preview any historical version
- Click Use This Version to revert to a previous version
Send to Movie Editor
Once your shots have generated videos, you can send the entire storyboard to the Movie Editor for final composition:
- Click Send to Movie Editor in the storyboard toolbar
- All completed shots with video are assembled into a timeline
- Fine-tune transitions, timing, and final output in the Movie Editor
Troubleshooting
Video Doesn't Match Hero Frame
- Try a different video model (some handle I2V better than others)
- Reduce motion intensity in the shot's cinematography
- Add more specific motion descriptions in the prompt
Artifacts or Glitches
- Regenerate (results vary between generations)
- Try a higher quality model (Veo 3 for best results)
- Simplify complex elements in the shot
Audio Issues (Veo 3 / Kling 2.6)
- Audio is generated natively by the model, not post-processed
- Regenerate if audio doesn't match the visuals
- For shots requiring specific audio, generate without audio and add audio separately
Duration Mismatch
- Check the model's duration constraints in the table above
- The system auto-clamps to the nearest valid duration
- If exact timing is critical, choose a model with matching constraints (Wan 2.5 has the most flexible range: 3-10s)
Best Practices
1. Start with Critical Shots
Generate important shots first (hero moments, character close-ups, key story beats) before committing to batch generation.
2. Test Before Batch
Generate 1-2 representative shots to verify model choice and settings before generating all shots.
3. Use Budget Mode for Drafts
Enable Budget Mode (Kling 2.6 Pro instead of Veo 3) during iteration, then switch to premium for final versions.
4. Monitor Credits
Video generation is the most credit-intensive operation. Review hero frames thoroughly before approving for video.
5. Plan for Regeneration
Not every shot will be perfect on the first try. Budget for some regeneration.