Video extend continues past the last frame. A retake re-rolls a span you marked — the last two seconds, a flubbed gesture, a face that melted — and stitches the new sample back onto the frames you kept. Together they are how you live inside a model's short duration ceiling without throwing away a good opening every time the ending fails.
The join is the job. An extension that starts mid-motion inherits velocity; an extension that starts on a held frame has nothing to continue and will invent a new action. A retake that changes the seed, the prompt, or the model is not a retake, it is a cut to a different clip. Seed-lock the parent. Keep ratio, resolution, and motion level. Colour-grade the chain once at the end, not per segment.
Drift still compounds. Two honest extensions can hold a face; five in a chain rarely look like the same shot. Past that, storyboard a new setup and cut. Duration is a menu, not a slider, and extend-retake is how you assemble past the menu without pretending one generate was a thirty-second film.
In practice
- Extend from a frame that is still moving. A dead hold gives the model nothing to continue.
- Lock the seed, ratio, and model before a retake, or you are generating a different clip.
- Stop chaining after two or three joins. A new shot is cheaper than a fifth drift.
The mistake to avoid
Retaking with a rewritten prompt and calling the mismatch a model bug. The span you kept and the span you re-rolled are no longer the same brief.
Where you will run into it
- Extend a Video's Length — Your clip ends too soon. Keep it going.
Related terms
Video extend
Video extend continues an existing clip past its final frame, generating new footage that starts from where the old footage stopped.
Seed lock
A seed-lock freezes the random seed across runs so prompt or setting changes are the only variable — the honest way to iterate a nearly-right generate.
Duration
Duration meaning: how long a generated clip runs, chosen before generation from the lengths the model supports, not trimmed after.
Temporal consistency
Temporal consistency is how well a generated clip keeps things the same from one frame to the next, with no flicker, shimmer or drifting detail.
Text-to-video
Text-to-video is generation from a written prompt alone — you describe a shot, the model invents every frame of it, and no image or footage goes in.
The all-in-one AI studio for creators. 60+ models for video, image, voice, music and lipsync, in your browser or on your phone.
Free on the web and on iPhone, one account.