manga-ocr

Why we included this project

Manga OCR targets the cases that trip up general-purpose Japanese OCR: vertical and horizontal text, furigana, unusual fonts, and scans that have seen better days. Its end-to-end model reads an entire text bubble in a single pass, so you do not have to split multi-line dialogue into lines beforehand. It installs with pip and ships a pretrained model, and it already powers two other projects, the Poricom reader and mokuro, which suggests it holds up outside a demo. If you are building anything around Japanese comics, whether a translation workflow or a reading setup, this is a practical place to start rather than a research toy.