kha-white/mokuro

What it solves

It allows Japanese learners to read manga in a browser with selectable text, enabling the use of pop-up dictionaries like Yomitan for easier comprehension.

How it works

It uses a pipeline of text detection and OCR (Optical Character Recognition) to process manga pages offline. It leverages comic-text-detector for finding text and manga-ocr for converting those images of text into actual characters. The results are stored in a .mokuro file (or legacy HTML files) containing OCR results and metadata, which are then loaded into a web reader to display the text overlays on the images.

Who it’s for

People learning Japanese who want to read manga in their native format while having instant access to dictionary lookups.

Highlights

• Offline Processing: All OCR and text detection are performed before reading, so the reader experience is more fluid.
• Web-based Reader: Supports a dedicated web reader for managing catalogs of processed volumes.
• GPU Acceleration: Supports CUDA for faster processing of manga volumes.
• Flexible Output: Provides both a modern .mokuro metadata format and a legacy HTML format for backward compatibility.