Keeping an Anime Character the Same Across Twenty Shots

Anime is unusually unforgiving of character drift. The style is built on consistent, simplified shapes — which means a change that would be invisible in a photoreal render is glaring in a cel-shaded one.

Reference first, always

Generate one character sheet before you generate any scene: front, three-quarter, profile, at consistent scale. Everything afterwards references that sheet. A text description alone — however detailed — will drift, because every generation re-interprets the words.

What actually drifts

In order of how quickly a viewer notices:

1. Hair silhouette. The outline, not the colour. Anime characters are recognised by shape at a distance, and a changed fringe reads as a different person.
2. Eye shape and spacing. Subtle and devastating.
3. Costume details. Buttons, trim, the direction of a collar fold.
4. Proportion. Head-to-body ratio shifting between shots.
5. Colour. Last, and the easiest to fix in post.

Note that colour is at the bottom — most people spend their effort there and their character still reads as inconsistent, because the silhouette moved.

Describe silhouette, not adjectives

"Long silver hair" is four different characters. "Hair falling to mid-back, straight, parted left, two strands forward of the ears" is one. Anything you do not pin down, the model will invent, and it will invent differently every time.

Shots that break consistency

• Extreme angles. Looking straight up or down forces the model to solve a shape it has not been shown.
• Fast action with the face visible. Motion plus identity is two hard problems at once. Either the action is close and the face is out of frame, or the face is clear and the movement is contained.
• Group shots. Every additional character multiplies the chance of a blend. Two is manageable; four is a lottery.
• Radical lighting changes. Backlight and rim light rewrite the silhouette, which is exactly what you are trying to protect.

The workflow that holds

1. Character sheet. Approve it before anything else.
2. Key frames as images for each scene, using the sheet as reference. Cheap, fast, and this is where you catch drift.
3. Animate only the approved key frames.
4. Review the whole set side by side before assembling — drift is obvious in a grid and invisible in sequence.

That last step catches more problems than any prompt refinement. Twenty shots viewed one after another all look fine; the same twenty in a grid show you exactly which three are wrong.

Doing this in Katama: Cinema Lab

Everything above is a method. If you want the method without assembling it by hand every time, that is what Cinema Lab is for — the infinite-canvas studio at katama.ai/cinema.

It is built around the four steps this article keeps coming back to:

• Canvas — lay your references and key frames side by side. This is where drift becomes visible, because you are looking at the shots together rather than one after another.
• Board — the shot list. Each card is a beat, and the beats stay in order while you change what is inside them.
• Compose — the prompt for each shot, next to the frame it produces. Change one, see the other.
• Timeline — the assembly, with the durations you actually generated rather than the ones you meant to.

Where consistency comes from

The part that matters for repeatable results is not the canvas — it is what sits behind it. A locked reference set plus a saved prompt structure means the tenth video in a series is built the same way as the first. That is the difference between a good clip and a body of work that looks like it came from one place.

And once a sequence works, it does not have to be rebuilt by hand: Workflows chains the steps into one pipeline, and Autopilot runs that pipeline on a schedule. Same references, same prompt structure, same look — on Tuesday and again three weeks later.

Start in Cinema Lab when the job is more than one shot. For a single clip, the Video Studio is faster and there is nothing to keep consistent.