# How do you keep locations consistent in AI video?

**Short answer:** Treat each location like a character: give it its own reference, and for scenes with several camera angles, build it as a set of camera plates (the same empty room from four to six angles) and have each shot reference the plate that matches its angle. Keep time of day and weather in every shot's description, because night drifts to day.

URL: https://arc.katechon.technology/answers/how-to-keep-locations-consistent-in-ai-video
Updated: 2026-10-18

## Why places drift

Faces get the attention, but viewers notice when a room changes shape between cuts. Each clip is a fresh generation: a location described in words is rebuilt every time, with doors, windows and furniture moving. The reverse angle is the worst case, because the model has never seen the other side of the room.

## Camera plates

A camera plate is an empty reference image of a location from one specific angle. On [Dragon Rising](https://arc.katechon.technology/films/dragon-rising) we built 16 locations with 4–6 plates each, and every shot referenced the plate for its angle, so geography held between fast cuts. For [The Emperor's Pleasure Court](https://arc.katechon.technology/films/the-emperors-pleasure-court) we re-shot every location of a vertical series as 16:9 plates, so the widescreen film matches the vertical episodes.

1. Generate the location once as an establishing image and approve it.
2. Derive plates from it: front, reverse, two sides, a high angle. Keep them empty of people.
3. In each shot, attach the plate that matches the camera, plus the characters.
4. Write time of day and weather into the shot, not just the location.

## Drift we saw and the fixes

- Night scenes drifting to daylight: state night and the light source in every night shot.
- Props leaking into sets: a style rule about a red balloon put balloons into eight set sheets; we wrote "empty plate, no balloons" into set descriptions.
- Mixed media: painted backgrounds with photoreal people broke the look; one painting language for both fixed it.
- Too many sets: our simplest series used six to eight sets, so the same plates appear in nearly every shot.

## Cost

Plates are cheap insurance. In Arc each is a Seedream 5.0 Lite image at 13 credits, so five plates for a location are 65 credits, less than one 5-second 720p clip on Seedance 2.0 Fast (113 credits).

## Where Katechon Arc fits

Arc treats locations as assets with their own references and supports multiple plates per location, attached per shot.

Arc does not build a 3D model of a set; if you need free camera moves through a space, a 3D scene tool fits better.

## Related answers

- [How do you keep the same character across AI video clips?](https://arc.katechon.technology/answers/how-to-keep-the-same-character-in-ai-video): Give the model the same visual evidence every time: one approved reference image per character, attached to every shot that character is…
- [How do you keep costumes and props consistent in AI video?](https://arc.katechon.technology/answers/how-to-keep-costumes-and-props-consistent-in-ai-video): Write the state of every costume and prop into every shot (helmet on or off, sword in hand or sheathed), give each distinct look its own…
- [What is a grid storyboard in AI video?](https://arc.katechon.technology/answers/what-is-a-grid-storyboard): A grid storyboard is one generated image divided into cells, each cell the frame for one shot, so several consecutive shots are drawn…
- [Should an AI drama be vertical (9:16) or horizontal (16:9)?](https://arc.katechon.technology/answers/vertical-or-horizontal-for-ai-drama): Vertical 9:16 for short dramas watched in phone feeds and short-drama apps; horizontal 16:9 for films and long-form YouTube.