Character consistency
Character consistency means the same figure looks the same across several shots: same face, same clothes, same hair, same build. It is the hardest technical problem in generative video production and the point where amateur work shows fastest.
Why this is difficult
Every shot is produced in its own computation. The model has no memory of what it produced a minute earlier. A description such as "woman in her mid-thirties, dark hair, blue jacket" permits a thousand different faces, and any of them can appear on the next run.
In the edit that is immediately obvious. Viewers read faces better than any other subject, and two slightly different faces do not look like an error, they look like two people.
How we establish it
- Reference images: a fixed image of the figure goes into every generation.
- Fixed descriptions: the same wording for appearance and clothing every time, not reformulated.
- Same seed: where the model allows it, as an additional anchor.
- Working from an image: image-to-video rather than text alone.
- Clean-up: inpainting where the face drifts.
What that means for your project
The longer a figure is on screen and the closer the camera, the more effort it takes. A film following one person through ten shots is noticeably more expensive than one with changing subjects. We say that in the first call, before it appears in the offer.
When a particular real person has to appear, filming those passages is usually the better route. The world around them can still be generated.
Frequently asked questions
What is character consistency?
Character consistency describes a generated figure looking identical across every shot of a film. It is established through reference images, fixed descriptions and clean-up, because generative models have no memory between two runs.
Does this work for a company mascot?
Yes, and usually better than for realistic humans, because a drawn figure has fewer details where deviations show. References from several angles help considerably.
Can a figure stay the same across several projects?
Yes. We keep a reference pack for that: images from several angles, the exact description and the settings used. The figure then looks the same in the next film as in the first.
Related terms
Read on: All terms in the glossary • AI video or traditional shoot
A figure carrying your film?
We tell you in advance what that means in effort and where the limit sits.
Book a first call