Back to selected work

Manga Translator & OCR

project · 2025

Manga Translator & OCR — case study

My role
Developer — manga translation and Japanese OCR integration.

The problem

Read Japanese manga without waiting for an English release.

The constraints

Text must first be located inside page artwork. Recognition errors then carry into translation.

Approach and tradeoff

The prototype separates speech-bubble detection, manga-ocr recognition and translation behind a FastAPI server. Separate stages are easier to inspect, but each can introduce errors. The README distinguishes its demonstrated YOLOv11 output from the planned YOLOv12 detector.

  1. Detect speech bubbles
  2. Recognise Japanese text
  3. Translate to English
Pipeline documented in the repository. Each stage can be inspected separately.

The result

A documented end-to-end prototype with a /predict interface. Translation quality and OCR fine-tuning gains are not quantified.

Inspect the evidence