manga-ocr (kha-white/manga-ocr): OCR for Japanese text, focused on manga
Source: https://github.com/kha-white/manga-ocr
Related: kha-white/mokuro (GPL-3.0, uses manga-ocr to make a selectable-text HTML overlay for manga; pushed 2026-07-20) is queued, not recorded. Hugging Face model: kha-white/manga-ocr-base.
Structured fields
{
"archived": false,
"revision": {
"ref": "v0.1.16 (latest release, published 2026-07-19); Hugging Face model kha-white/manga-ocr-base lastModified 2022-06-22",
"pinned": true,
"source_url": "https://github.com/kha-white/manga-ocr/releases"
},
"checked_at": "2026-10-17T00:00:00.000Z",
"usefulness": "Vision-Encoder-Decoder OCR model and Python package for Japanese text in manga: vertical and horizontal text, furigana, overlays, many fonts and low-quality images, multi-line speech bubbles in one pass. Output is plain text for translation or dictionary workflows; its model was trained with Manga109-s plus synthetic data, which affects reuse.",
"source_type": "repository",
"availability": "available",
"dependencies": [
"Python 3.9 or newer",
"PyTorch",
"transformers",
"fugashi / MeCab (per troubleshooting notes)"
],
"source_owner": "kha-white (individual developer); location not verified",
"next_check_at": "2026-10-24T00:00:00.000Z",
"hardware_notes": "GPU optional: install PyTorch with CUDA as described on the PyTorch site, otherwise CPU; no VRAM figures given.",
"known_limitations": [
"Included because the lead named it for Japanese manga OCR; the developer's country was not verified and the README is English, so it is a tool for the Japanese ecosystem rather than a Japan-published source.",
"The weights are Apache-2.0 on Hugging Face but the training data included Manga109-s, whose terms require disclosing use when publishing pre-trained models; the README acknowledges the dataset but does not discuss Manga109-s terms.",
"No benchmark numbers read."
],
"upstream_updated_at": "2026-07-19T08:43:43.000Z"
}Evidence claims
- [sourced_fact] The README describes optical character recognition for Japanese text with the main focus on Japanese manga, using a custom end-to-end Vision Encoder Decoder model. · locator: README intro source
- [upstream_claim] Unlike many OCR models it supports recognising multi-line text in a single forward pass so speech bubbles can be processed at once. · locator: README intro source
- [sourced_fact] The README acknowledges that the project used the Manga109-s dataset and the CC-100 dataset. · locator: README, Acknowledgments source
- [sourced_fact] The LICENSE file is the Apache License 2.0. · locator: LICENSE source
- [sourced_fact] The Hugging Face model kha-white/manga-ocr-base is tagged license:apache-2.0 and dataset:manga109s. · locator: API tags source
- [sourced_fact] GitHub API returns the repository record for kha-white/manga-ocr: default branch master; pushed_at at check time 2026-07-19T08:43:43Z (raw response saved under evidence/github/kha-white--manga-ocr.json). · locator: GitHub API repos/kha-white/manga-ocr, default_branch (pushed_at is volatile) source
- [sourced_fact] GitHub API reports archived = false for kha-white/manga-ocr. · locator: GitHub API repos/kha-white/manga-ocr, archived source
- [sourced_fact] GitHub license detection for kha-white/manga-ocr returns spdx_id Apache-2.0. · locator: GitHub API repos/kha-white/manga-ocr, license source
- [sourced_fact] GitHub API latest release for kha-white/manga-ocr: tag v0.1.16 published 2026-07-19T08:44:54Z (selected fields of the response saved under evidence/github/kha-white--manga-ocr.release.json). · locator: GitHub API repos/kha-white/manga-ocr/releases/latest, tag_name source
- [upstream_claim] Research review verdict at export: reviewed (as of 2026-10-17). Recorded by the research pipeline, not a hub review outcome; evidence.review stays unreviewed until an independent reviewer records one.
Rights
| component | license | hosting | redistribution | modification | training | Evidence and notes |
|---|---|---|---|---|---|---|
| code | Apache-2.0 | permitted | permitted | permitted | unknown | Apache-2.0; notice and patent terms apply. Does not address training. |
| model_weights | Apache-2.0 | permitted | permitted | permitted | unknown | Hub license tag says apache-2.0; the model was trained using the Manga109-s dataset (tag dataset:manga109s), whose terms require the use of Manga109-s to be indicated clearly when publishing pre-trained models and forbid redistributing the images; compliance of the published weights with those conditions was not assessed. |
Linked records
- Does Manga109-s permit training generative image or text-to-speech models, and does the manga-ocr-base release satisfy the disclosure condition? discusses → this
- Method (conditions as written on the dataset website). 1. Choose the subset: Manga109 (109 volumes) is limited to academic, non-commercial purposes by non-commercial organisations; Manga109-s (87 volumes) is approved for commercial use. 2. Never redistribute images or parts of the dataset to third … cites → this
- The Hugging Face tags for kha-white/manga-ocr-base show license:apache-2.0 and dataset:manga109s; the README acknowledges Manga109-s and CC-100. Manga109-s terms require that published results, including pre-trained models, clearly indicate use of Manga109-s. The README text read does not discuss t… cites → this