Manga & Anime Style Book
Search styles 日本語

API guide

The styles on this site can be read as JSON by programs. Read-only; no sign-up is needed.

Two URLs

GET https://manga.akiba.tokyo.jp/v1/tree

Every published branch (levels 1 to 3), each with id, parent (the parent id; null at level 1), level, title, summary and thumbnail_url.

GET https://manga.akiba.tokyo.jp/v1/styles/{id}

One branch: its prompt, its sample image and the branches one level below (children).

Add ?lang=en for English names and descriptions (Japanese by default). Prompts always carry both Japanese and English.

Choosing a style

  1. Level 1 is the branches of /v1/tree whose parent is null.
  2. Pick a level-2 branch from children of /v1/styles/{level-1 id}, then read /v1/styles/{level-2 id}.
  3. Pick a level-3 branch from its children. You may stop at any level (the levels below are left open).

Using it for image generation

  • Use prompt.en (or prompt.ja) as the style and append the person and scene you want. The prompt has no subject.
  • avoid lists what must not appear: text, speech balloons, sound-effect lettering, panel borders, signatures, watermarks and colour.
  • The sample image samples[reference].url can be passed as a style reference image.
  • A legacy level-3 prompt is the level-2 words + "Finished style (where it differs from the above, follow this): " + the level-3 words. Where they disagree (amount of black, head-to-body ratio and so on), follow the level-3 words.

From Genko Studio

A branch page opened with ?client=genko shows "Use this style", which opens genko://use-style?id={id}. In an ordinary browser you can copy the branch's API URL instead.

Other notes

  • HTTPS, readable by anyone. CORS allows GET only.
  • When the API is limited, send Authorization: Bearer <token> (401 without it).
  • Unknown or unpublished branches return 404 ({"error": "not found"}). Methods other than GET return 405.
  • Legacy samples are 768×1024 JPEG. Explanatory pairs retain their aspect ratio at up to 1024 pixels on the long edge: L for monochrome and RGB for color branches. Their 384×512 thumbnails are padded, not cropped.
  • When a sample image or the words change, version goes up and updated_at changes.

Fields of nodes[] in /v1/tree

id
Branch id (lowercase letters, digits and hyphens).
parent
Parent branch id; null at level 1.
level
Level (1 to 3).
title
Display name.
summary
One-line description.
thumbnail_url
Thumbnail of the branch's sample (384×512); null without a sample.
path
Display names from level 1 down to this branch (this branch last), e.g. ["Gag manga", "Surreal", …]. The same name exists under other categories, so never pick a style by its name alone.

Fields of /v1/styles/{id}

id
Branch id.
path
Ids and names of the levels above (from level 1 down to the parent), top first; an empty list at level 1.
level
Level (1 to 3).
title
Display name.
summary
One-line description.
expression
"mono" is black-and-white and "color" is full color. Existing branches stay mono.
model
id is "chatgpt-image"; version is the image model that made the sample (null without a sample).
prompt
ja, en and tags: the common words + the words down to this branch (level 1 alone; from level 2 down otherwise). No subject.
avoid
What must not appear in the picture.
samples
Sample images: one legacy upper_body, or representative followed by application for an updated branch. Each has subject, url and prompt (the actual Japanese and English generation text).
reference
Index in samples of the best sample for the style (from 0); null without a sample. Not a count of reference images.
license
generated is always true, artist_names always false; terms are the conditions of use.
version
Goes up when the sample image or the words change.
updated_at
When it last changed (UTC).
children
Published branches one level below (id, title, summary, thumbnail_url); empty at level 3.
comparisons
Comparison pictures (added 2026-10-18): the prompt that made the sample (samples[0].prompt.en), unchanged, drawn by another image model. model (id, version, name), url, prompt_from, settings (seed, steps, cfg_scale, sampler, size, negative_prompt) and created_at. Not the style (use samples[reference] as the reference image). An empty list when there is none; adding one does not change version.

Live responses