Writing a shot list the models can actually render
Durations that exist, coverage that cuts, and the four constraints worth knowing before you plan a scene rather than after.
- Written by
- The Cetus team
- Seeds a claim
- 5
- Models
- 3

On this page5
A shot list written for a camera crew and a shot list written for these models are close, but not the same, and the differences are all in the same direction: the models are more literal about time and less literal about space.
Durations are a menu, not a slider
Omni Flash renders 4, 6, 8 or 10 seconds. Veo 3.1 Lite renders 8, and only 8 — a request for 5 is refused, not rounded. Everything about planning follows from that.
- A twelve-second shot is two shots. Plan it as two, with a cut point you chose, rather than discovering at render time that it has to be two and cutting wherever the failure happened.
- Do not plan a scene that only works at one exact length. If the beat needs 7 seconds, it needs 6 or 8, and 6 is usually better.
- A held moment is a shorter clip on screen for longer, not a longer clip. Four seconds of stillness reads as stillness; ten seconds of a model trying to hold stillness reads as drift.
Write coverage, not masters
The instinct from writing for a camera is to plan a master and pick from it. There is no master here — every shot is a separate generation with its own risk of failing — so a scene should be planned as the pieces you would cut to anyway.
A two-person conversation, planned for these models, looks like this:
1 WIDE 8s both at the table, static
2 SINGLE A 6s she speaks, slow push in
3 SINGLE B 6s he listens, static
4 INSERT 4s her hand turns the cup
5 SINGLE A 6s she finishes, static
6 WIDE 8s he stands and leaves frame right
38s of coverage. Six generations. Any one can
fail and be re-run without touching the others.Compare that with one 40-second two-shot, which no model in the catalogue can produce and which, if it could, would be one failure away from nothing.
One instruction per axis
Each shot gets exactly one shot size, one camera instruction, one light description and one speed. Two camera instructions in a sentence — slow push in as the camera drifts left — produce neither reliably, because the model has to decide which one you meant and it does not tell you what it decided.
If a shot needs two camera moves, it is two shots. That is also true with a real camera; it is just cheaper to find out here.
The four constraints, written down
- Time. 4/6/8/10 on Omni Flash, 8 on Veo 3.1 Lite. Plan against the menu.
- Identity. Anything recurring needs a reference still. Budget the stills into the list rather than adding them later.
- Contact. Hands touching objects, pouring, handing over — these are the shots that fail. Put them on Veo 3.1 Lite, keep them short, and shoot an alternative.
- Room. Models will not invent space that is not in frame. If something enters, leave the space it enters from.
Order the list by risk, not by story
Render the hard shots first. A scene whose one difficult insert turns out to be impossible is a scene you want to rewrite before you have made the other nine shots around it, not after.
In practice that means: reference stills, then contact shots, then anything with two people, then everything else. It is the reverse of shooting order on a real set, and it is right here because re-rendering is cheap and re-planning is not.



