Scene Generation
Scene generation is the production phase where AI creates a still image and then animates it into a video clip for every scene in your project.
Two-Stage Generation
Still Frame
Stage 1: Image Generation
For each scene, a still image is generated using the art direction image prompt. The frame becomes the starting keyframe for video generation.
Which model draws that frame is the Image Model you picked for this project on its setup page — a per-project choice, so one project can use an Azure GPT Image model and another can use Nano Banana. A project keeps the model it is set to; a model added to the list later never changes a project already in flight.
Reference images (if uploaded) are used to maintain visual consistency across scenes.
Animation
Stage 2: Video Generation
Each scene image is animated into a video clip using the art direction video prompt and your chosen video model. The video clip length is automatically matched to the scene's planned duration based on available clip lengths.
Duration Bucketing
Each video provider supports specific clip durations. The platform automatically selects the best-fit bucket for each scene:
| Provider | Available Durations |
|---|---|
| Veo 3.1 | 4s, 6s, 8s |
| Sora 2 | 4s, 8s, 12s |
| Gemini Omni 1.1 Flash | 3s–10s |
| Runway Gen4.5 | 5s, 10s |
| Luma Ray-2 | 5s, 10s |
If the actual video duration doesn't match the planned duration (within tolerance), the scene is flagged as duration incompatible and can be repaired.
Scene Recovery Options
When a scene fails or produces unsatisfactory results, you have several recovery options:
Retry Video
Keep the existing image and regenerate only the video clip. Useful when the image is good but the animation didn't work well. You can switch to a different video provider for the retry.
Regenerate
Regenerate both the image and video from scratch. The art direction prompts are preserved, but new outputs are generated. The recovery options offer this per image model, so you can redraw one scene with a different image model without changing the project's own setting or touching any other scene.
Remix
Create a variant of the existing video using Sora's remix capability. Produces a visually related but different interpretation of the same prompt.
Edit prompt & retry
Rewrite the scene's image or video prompt yourself and retry. The edit is saved to the project's storyboard, so a Producer retry uses your wording. This is the way out when a provider's safety filter rejected every automatic rephrase.
Regenerate with custom prompt
Regenerate the still from a prompt you write, keeping your project's reference images, then re-animate it. Useful when the filter objected to the image rather than the words.
Repair Incompatible
Fix scenes where the actual video duration doesn't match the planned duration. Re-generates the video with adjusted duration parameters.
Batch Retry Failed
Retry all failed scenes at once. The platform analyzes each failure and builds appropriate retry prompts automatically.
Automatic Retries
When video generation fails, Filmgenic automatically classifies the issue and suggests the best retry approach. Common scenarios include safety filter triggers, timeouts, and temporary service issues.
Each scene supports up to 3 retry attempts, and a rejected prompt is rephrased automatically before an attempt counts as failed. Temporary provider limits (rate limits or quota) are retried with a short wait and never consume an attempt. Generation history is preserved so you can see what was tried.
Locks and takes
Once a shot is right, lockit. A lock pins the scene to its current take: Retry, Re-render, Regenerate, Remix, recovery actions and prompt edits are refused until you unlock it, and the Producer's review will not replace a locked shot on its own. If a locked shot fails review, the run pauses for your decision instead of regenerating it.
Per scene
Lock / Unlock
Version history
Takes
Scoped & costed
Revise a shot
No regeneration