koharu

Koharu is a local-first, ML-powered manga translation tool written in Rust. It automates the translation process by combining object detection, OCR, inpainting, and LLM capabilities to deliver a seamless workflow. The software runs vision models and LLMs locally to ensure data privacy and security. Key features include multi-format project management (raster images, archives, PDFs), selective pipeline processing (detection, OCR, translation, inpainting), and support for text regions, speech bubbles, and cleanup. It offers multimodal OCR, generative inpainting for source-text removal, and proofreading tools. The editor includes a WebGPU-based canvas for manual cleanup, text placement, and page composition, with multilingual text shaping (CJK, RTL) and layered PSD export. Koharu supports hardware acceleration via CUDA, ROCm/HIP, Metal, Vulkan, and WebGPU. It integrates with local GGUF models (e.g., LFM 2.5, Ministral 3, Gemma 4, Qwen 3.x) and hosted providers (OpenAI, Gemini, Claude, DeepL, etc.). Available for Windows, macOS, and Linux.