Anime is a style with rules: flat cel shading, clean line work, expressive eyes, dramatic skies. Prompts that respect the rules get the look; prompts that mix in photographic language do not. We have generated enough anime frames to know which words hold across thirty scenes and which fall apart by scene four, and the short version is that the word "anime" on its own is the least useful word in the prompt.
What the model actually knows
Image models have seen millions of anime frames, tagged inconsistently, spanning decades of cel animation, modern digital production, merchandise art and Western imitations. The word alone lets the model pick any of a dozen sub-styles, and it will pick a different one for every scene. Naming the conventions narrows the choice. Naming a specific studio is unreliable and, for a commercial product, unwise.
Words that hold
"Cel shading", "clean line art", "flat colors", "anime film still", "wide shot", "large expressive eyes", "soft painterly background", "warm rim light". Add a time of day and a weather; anime lives on skies and light. Add a colour note, such as a muted teal and amber palette, to unify scenes, and a camera note to unify shots. Keep the style sentence under forty words and put it first in every prompt; models weight the beginning.
For video prompts, add motion vocabulary from the medium: limited animation, held frames, a slow pan, hair and cloth moving in the wind. A video model asked for anime without motion cues produces a photoreal camera move over a drawn frame, which reads as a poster being filmed.
Words that break it
"Photorealistic", "8k", "depth of field", "skin texture". These pull the model toward photography and you get an uncanny hybrid: a drawn face with pores. The same creep happens without any prompt help on the cinema video tiers, which add texture and lighting that flat styles do not want. For anime, the faster tier is usually the better one, and it costs less.
Composition
Anime frames are often asymmetric: the character to one side, the world filling the rest. Ask for it. "Character at left third, city skyline right" is a real instruction for the image model, which honours spatial language far better than a video model does. Compose the frame first; the video model only has to keep it.
Backgrounds deserve their own sentence. Anime treats them differently from characters: painterly, detailed, static. A prompt that says "painterly background, cel-shaded character" gets the contrast right; one that says only "anime" often produces a cel-shaded background that looks like a game.
Traps we hit
- Decorative borders. The model draws a frame line or a panel edge. Say "fills the whole image edge to edge, no border" and check the keyframe before animating; a framed keyframe becomes a framed clip.
- Text and captions. Anime frames in training data carry subtitles and titles, so the model adds lettering. Forbid text explicitly; it cannot be removed after animation.
- Humanising the non-human. A firefly becomes a winged girl unless the written look says "an actual firefly, an insect" and the cast sheet shows one. The training data is full of humanised animals.
- Mixed proportions. Without a reference sheet, one scene draws the character chibi and the next realistic. Sheets fix it; the style sentence alone does not.
Keep the character in the bible
Style consistency across scenes comes from the bible as much as from the prompt. Build the character's references in the anime style you want, and every scene inherits it. Consistency is a pipeline property: the same sentence in thirty independent text-to-video calls still yields thirty slightly different looks, because each call re-samples. Sheets drawn once, keyframes composed from the sheets, clips animated from the keyframes: that is what pins the look.
Choosing the tier
Draft and standard tiers animate the keyframe with the fast video model, which keeps flat colour flat and suits the five-second clips that limited animation likes; holds and cuts are part of the language. Reserve the cinema tiers for a scene that needs complex motion, such as water or a crowd, and accept some texture creep there. Story to film offers Anime as one of its eight styles, and the style guide it writes is repeated into every cast sheet and scene prompt automatically.
Other drawn styles
The same method transfers. Watercolor: wet edges, paper grain, soft bleeding colour, no line. Ink wash: monochrome brush strokes, negative space, one accent colour. Comic book: bold ink outlines, halftone, flat saturated colour, no speech balloons. Pastel: chalky shading, rounded shapes, a warm palette. In every case, name the medium's conventions, forbid text and frames, fix the sentence, and let sheets and keyframes carry it.
Frequently asked
Why does the model keep adding a caption? Because anime frames in its training data often carry one. "No text, no caption, no subtitles" at the end of the prompt is usually enough; if not, regenerate the keyframe before animating.
Can a photo be turned into an anime character? Yes, as a reference for the cast sheet, with the person's consent. Generate the sheet in the style from the photo, then use only the sheet downstream so realism does not leak.
Do negative prompts work on the video model? Inconsistently. Put the forbidden items in the keyframe prompt, where the image model honours them, and keep the video prompt about motion only.
