A character reference is only the start of an AI video brief

A reference image answers what a character looks like. A shot still needs a job: open the door, react to someone, cross the room. Then it has to fit the next shot.

Here is a review exercise to try before paying for another generation:

1. Write one sentence describing the action the shot must communicate.

2. Mark the usable start and end of the generated clip. Judge the movement, not just its best frame.

3. Watch that section beside the neighbouring shot. Check whether the character, direction and action still make sense.

4. Record a decision: keep, trim or regenerate, with a specific reason. A problem outside the usable section may only need a trim.

For example, a character might look right while standing still but change appearance during a turn. A shorter usable section may work in the edit; if the turn is the point of the shot, a trim does not solve the brief. This is an illustrative review exercise, not a claim that a particular tool can guarantee the result.

My own production cases include roughly 30 AI idol-video versions and seven musical-style PLANET ME videos. I keep the idol motion-review cases and musical character-planning cases separate in my AI Music Video Production Toolkit.

The English toolkit is USD $49: two visual guides totalling 49 pages, 16 silent study clips and 11 editable CSV worksheets. The clips are private study material, not stock footage or finished videos for reuse. No generation credits or software subscriptions are included.

Contents and previews: https://lucete.gumroad.com/l/check-in-shot-selection?utm_source=indiehackers&utm_medium=product_update&utm_campaign=shot_review_20260930

I run Lucete AI Lab and sell this toolkit. The production examples are my own; AI assisted the English editorial preparation.

When you review an AI-generated clip, what most often changes your decision after you put it next to another shot?