Separate a scene from a shot
A scene is a story event in a place and time. A shot is one continuous camera setup within that scene. A conversation can be one scene and several shots. Your plan becomes easier to generate and edit when those two levels are clear.
For a short AI film, describe one main action per shot. A character arriving, opening a vehicle, discovering someone and reacting may need several shots even though the whole event happens at one location.
Write the start before the movement
Describe the actual first moment, not a summary of the whole scene. Name who is visible, where their hands are, what they hold and whether doors or other moving parts are open. Then describe the action that changes that arrangement.
An ending is still useful written direction. It does not require a second generated image. The video can begin with one approved opening image and move toward the final state you describe.
Assign references a clear role
A character reference establishes appearance. A location reference establishes the place. An earlier shot image can help continue an arrangement, but it should not freeze the next shot in the wrong pose or force every person from the previous shot to appear.
Name who owns each important prop. If one person carries a gun, a phone or a bag, say so explicitly and keep it out of other characters’ hands. Review this in the opening image before generating motion.
Budget time for the audience
A recognition beat needs time to read; a simple detail shot may be brief. Add your planned durations and watch the assembled cut before deciding that the total is right. Generated clips can contain useful extra footage that you do not need in the edit.
Read dialogue aloud within the planned shot length. If it only fits at a rushed pace, shorten the line or give it more time. Include room for a response when the response is part of the story.
Use the list as a review checklist
- Check every approved image against the opening pose and reference list.
- Watch the motion to confirm the requested action, prop ownership and final moment.
- Listen for the exact spoken line and unwanted speech or music.
- Play neighbouring shots together to check position, lighting and sound continuity.
- Make one focused revision when something fails, then review that shot again before the full cut.