From Generated Clip to Shot List: Using AI Video as a Pre-Visualisation Tool

The most underrated use of AI video has nothing to do with publishing AI video. It is pre-visualisation: generating a rough version of a shot so that a real crew, a client or a director can argue with something concrete instead of a paragraph.

Production people already do this with storyboards and animatics. Sketches are fast but vague about time. Animatics fix the timing but look like cardboard. A generated clip sits between them: rough enough to change, specific enough to show how a camera move actually feels over two seconds.

The workflow is straightforward.

Start from the shot list, not from a mood. Each line already describes a subject, a frame and a move. Feed those into the generator one at a time, and keep the output labelled by shot number so the set stays navigable.

Generate two versions of any shot that is expensive to shoot. If both versions read clearly, the shot is safe to schedule. If neither does, the idea is the problem, not the execution, and you have saved a day on set.

Cut the clips against the scratch audio. Timing problems surface immediately: a beat that needs another half second, a transition that does not read, an ending that lands late.

Then take the surviving clips on set. A generated clip is a better brief than a paragraph because it settles arguments about energy and pace. Crews respond to it in a way they do not respond to words: they can see what "slow push in" means here.

The honest limitation: generated clips are not reference for lighting or lens choice. Treat them as blocking and timing reference, and let the DP do their job.

Wan 3.0 makes that pre-vis loop cheap, because you can iterate eight versions of a shot before lunch, and the free tier covers the first round: https://wan-3.io/

Cheap shots make expensive shoots better.