Summary This Android application automates manga translation using LLMs, offering instant, overlaid translations and screen-wide capabilities for any app. It benefits readers by integrating local OCR with AI-driven text processing for an enhanced comic-reading experience.

How It Works The pipeline utilizes local ONNX models for speech bubble detection and OCR (Japanese, English, Korean). Detected text is sent to an OpenAI-compatible API for LLM translation into Chinese (Simplified/Traditional). Translated text appears as draggable bubbles overlaid on original manga images.

Quick Start & Requirements Building requires JDK 17.0.17+, Kotlin 2.0.0+, Gradle 8.11.1+, and Android SDK platform 35. Developers must download and place specific ONNX models (e.g., comic-speech-bubble-detector.onnx, OCR models) into assets/. Configuration involves an OpenAI-compatible API endpoint, API Key, and model name. A tutorial is available at https://github.com/jedzqer/manga-translator/blob/main/Tutorial/简中教程.md.

Highlighted Details Features include screen translation via a floating window for real-time OCR/translation on any app. A manga library manager supports CBZ, ZIP, PDF imports, batch operations, and per-folder glossaries (glossary.json) for consistent term translation. It offers standard and "full-text fast translation" modes.

Maintenance & Community Community interaction occurs via a QQ group (ID: 1080302768). The project acknowledges PaddleOCR and kha-white/manga-ocr. Specific maintainer details or sponsorships are not provided.

Licensing & Compatibility The README omits a software license, creating ambiguity for usage rights, modification, and distribution, especially for commercial use. Compatibility is primarily for Android.

Limitations & Caveats Translation accuracy and order depend on correct image file naming. The API endpoint must use the /v1 suffix, and network connectivity is mandatory for LLM translation. Users must ensure API provider and model name match configurations.