Koharu is an AI-powered manga translator built in Rust, designed to automate the entire translation workflow. It targets users seeking an efficient, seamless way to translate manga by leveraging advanced AI models for text detection, recognition, removal, and translation, offering significant speed and safety benefits.

How It Works

Koharu employs a multi-stage AI pipeline: object detection for speech bubble segmentation, OCR for text extraction, inpainting to remove original text, and LLMs for translation. It utilizes Rust for performance and safety, with candle enabling high-speed, local AI inference and Tauri providing a cross-platform GUI. This integrated approach streamlines manga translation, handling complexities like vertical CJK text.

Quick Start & Requirements

• Installation: Download pre-built binaries for Windows and macOS from the releases page.
• Prerequisites: Rust (1.85+), Bun (1.0+).
• GPU Acceleration: Optional but recommended. Supports NVIDIA GPUs (CUDA compute capability 7.5+) and Apple Silicon (Metal). Bundles CUDA toolkit 12.x and cuDNN 9.x.
• Links: Discord Server

Highlighted Details

• AI Models: Integrates comic-text-detector, manga-ocr, and AnimeMangaInpainting models, converted to safetensors for Rust compatibility.
• LLM Support: Leverages candle for quantized LLMs in GGUF format (e.g., vntl-llama3-8b-v2, sakura-galtransl-7b-v3.7), downloaded on demand.
• GPU Acceleration: Optimized inference via CUDA (NVIDIA) and Metal (Apple Silicon), significantly boosting processing speeds.

Maintenance & Community

The project maintains a Discord server for user support and community interaction. Sponsorship is encouraged via GitHub Sponsors and Patreon to aid ongoing development.

Licensing & Compatibility

The Koharu application is licensed under the GNU General Public License v3.0 (GPLv3), imposing copyleft requirements. Its underlying sub-crates utilize the permissive Apache License 2.0. GPLv3 compatibility is crucial for any derivative or linked works.

Limitations & Caveats

Pre-built binaries are currently limited to Windows and macOS. Users on other platforms will need to build the project from source, which requires familiarity with Rust and Bun development environments.