Why the room changes after every cut in AI video
AI video models are good at faces and bad at rooms. In a multishot render the model sees your location once, from one angle, and rebuilds the set after every cut, so furniture moves. Give each beat its own start frame, one picture of how that camera setup opens, and the room holds across the cuts.
The sofa that moved
We rendered the same 22-second scene from Blood Child, our sample Nollywood micro drama episode, four times in ShowMaker Studio. Three takes had no start frames, and every one of them got the room wrong somewhere. In one, a character sat down on a sofa that wasn't there a cut earlier. The fourth take had a start frame for each beat, and the room held from the first cut to the last.
Why it happens
A multishot render is several shots in one clip, with cuts between them. The model is shown your location once, from one angle. After each cut it has to work out the new camera position and rebuild the set from scratch, and it guesses. Faces survive because the character reference keeps them anchored. The room doesn't have that anchor past the first shot.
The fix: one picture per beat
A start frame is a still of how a beat opens. Give every beat one, and the model doesn't have to invent the room after a cut. It's shown the room instead.
- Write one beat per camera setup. Each new angle is its own beat.
- Describe each beat's opening frame. Say who's where and what's in shot when that beat starts.
- Draw, pick or adjust a start frame. Drawing one costs 4 credits. You can also pick an existing frame and adjust it.
- Read the free room check. Before you render, check that each frame shows the same room.
- Audit the beats and voice the lines. Get the dialogue recorded before you spend on video.
- Render once. Make one multishot take for the scene.
- Check the take against its frames. Compare each beat of the render to the start frame it was given.
Habits that keep renders clean
- Fix the room in the stills before you get to video. A wrong chair in a 4-credit frame is cheap to fix. A wrong chair in a finished render means another render.
- Keep the location as a reference. A location with reference stills is cited as an @slug on every shot, so the same room is sent each time.
- Don't skip the opening frame on a reverse. A reverse angle shows a side of the room the model hasn't seen yet, which is where the guessing gets worst.
What it costs
Start frames add a few stills per scene at 4 credits each. In our test, three renders without them all came back wrong. One render with them held. Rendering video again almost always costs more than drawing the frames.
Watch the four takes: Multishot + start frames. For keeping a location steady across a whole episode, see AI location consistency, and the full guide at making a micro drama.
Related questions
Why do faces stay consistent when the room doesn't?
The character reference keeps a face anchored on every shot. In a multishot render the model sees the location once, from one angle, and has to rebuild the set after each cut.
How much does a start frame cost?
Drawing a start frame costs 4 credits. You can also pick an existing frame and adjust it.
Do I need a start frame for every beat?
In our test the take with a start frame for each beat was the only one where the room held from the first cut to the last. Three takes without them all got the room wrong somewhere.
Every fix on this page is a feature, not a workaround.
ShowMaker Studio is built around these problems — reference-locked casting, location look signatures, coverage-based scenes, delivery-timed captions, and per-shot model routing. Start free with 200 credits: narration over AI stills, no card.