Koharu

Koharu is an open-source, ML-powered manga translator written in Rust. It automates the full scanlation pipeline, from detecting text on a page to rendering the translation back into the artwork. The project has about 6,400 stars on GitHub, is dual-licensed under MIT and Apache-2.0, and is under active development with updates in 2026.

Overview

Koharu runs its vision models and language models locally on your machine, keeping your library private. It combines object detection for text regions and speech bubbles, OCR engines such as PaddleOCR VL and Manga OCR, generative inpainting with models like LaMa and FLUX to erase the original text, and LLM translation using local GGUF models or cloud providers such as OpenAI, Gemini and DeepL. A WebGPU-based editor canvas lets you proofread, adjust text placement and compose pages manually, with layered PSD export for further editing.

Key Features

• End-to-end manga translation: detection, OCR, translation and inpainting
• Local-first inference for vision models and LLMs, no cloud required
• GPU acceleration via CUDA, ROCm, Metal, Vulkan and WebGPU
• Support for local GGUF models plus hosted LLM and translation providers
• Speech-bubble and text-region detection with layout-aware rendering
• Proofreading tools and multilingual text shaping, including vertical CJK
• Layered PSD export and project management for images, archives and PDFs

💳 Pricing

Koharu is completely free and open source under MIT/Apache-2.0 dual licensing. There are no paid tiers or subscriptions; optional costs are only for any third-party cloud translation APIs you choose to plug in.

Who Is It For

Koharu is for scanlation groups, manga fans and localization hobbyists who want to translate manga on their own hardware with modern ML models instead of manual, page-by-page editing.