AI adaptation
Every capture scene starts as Keep original. If a capture would explain better as an animation, choose Allow AI adaptation, for one scene or all of them. That choice sends nothing and generates nothing.
Create a candidate
Choose Adapt with AI…, then pick a result:
- Animate with Qontent Tube returns a validated animation and needs no image model. This is the recommended choice.
- Generate media creates a new picture with an image model.
- Best fit uses animation when possible, otherwise media.
The dialog shows what will happen to each scene, and exactly what will be sent to which provider. Cancelling sends nothing.
A screenshot is sent as itself. A recording is sent only as stills you select, so the result doesn't transform its motion. Full recordings aren't sent to the currently supported providers.
No key? Choose to copy the prompt into any chat product, then paste the reply. Nothing is sent from Qontent Tube, and the reply gets the same validation.
Review it
Each result is a candidate:
- Compare previews it against the original.
- Accept selects it. It still can't render.
- Approve scene makes it the scene's output.
Regenerate adds another candidate, Reject sets one aside, and Use original returns to the capture. A candidate goes stale if its original or directive changes.
Why it stays safe
An animation is data, not code: text, cards, flows, particles, small trusted components, and simple 3D, moved by dialogue anchors and a fixed seed. Scripts, markup, stylesheets, shaders, and URLs are refused. Rendering uses the stored result and never calls a provider, so seeking, retrying, and re-mixing can't change it or cost anything.
Adapted scenes are labeled AI-adapted in Studio and in render metadata. Exact-evidence scenes can't use them.
Suggest framing
Suggest framing… drafts regions, camera, and overlays from one capture you agree to send. Nothing applies until you accept it. Its privacy hints don't count as a review.