Versions:
Koharu is an ML-powered manga translator written in the Rust programming language, published by mayocream and currently distributed at version 0.82.1 across a release history encompassing 44 versions. The software occupies the machine translation and computer-aided translation category, with a specific focus on comics and manga content, and it introduces a local-first workflow designed to automate the translation process through machine learning. Rather than treating translation as a single step, Koharu combines several distinct ML capabilities into one seamless experience: object detection is used to locate text regions within manga pages, optical character recognition (OCR) extracts the source text from those regions, inpainting removes the original text from the artwork so the page can be cleaned, and large language models (LLMs) handle the actual translation of the extracted content. By chaining these stages together, Koharu addresses the full pipeline that manga translation typically requires, from identifying speech bubbles and captions on the page to producing translated output that can be placed back onto cleaned artwork. The local-first design described by the publisher means the workflow is oriented around running these processes on the user's own machine, positioning Koharu as a tool for individuals and groups who want to translate manga without depending on a remote service for the core processing steps. Use cases center on automating the repetitive and technically demanding parts of manga translation, where detection, text extraction, artwork cleanup, and translation would otherwise each require separate manual effort or separate tools. Written in Rust, the project reflects a systems-language implementation for this ML-driven pipeline. With 44 versions recorded and version 0.82.1 as the current release, Koharu's version numbering indicates ongoing development prior to a 1.0 milestone, with the publisher mayocream maintaining the project as its feature set around detection, OCR, inpainting, and LLM-based translation continues to evolve.
Tags: