kirogii/mangaamtl

MangaAMTL is an open-source tool that automatically translates manga, primarily from Japanese, into other languages. It supports multiple source languages including Chinese, Korean, Japanese, Russian, and Indonesian, translating them into English. The author describes it as a personal project that is still in alpha testing, meaning bugs are expected and it should not be used for a public or production setting without understanding the risks involved. The tool runs as a local web application on your own computer. Setup requires Python 3.12, along with C++ build tools needed to install some of its dependencies. After downloading the latest release and installing the required Python packages, either a standard version or a version that uses your graphics card through CUDA for faster processing, you start the program and access it through a web browser at a local address. Under the hood, MangaAMTL relies on several existing AI models hosted on Hugging Face, a platform for sharing machine learning models. These include a language model for translation, a model for detecting where text boxes appear on a manga page, an optical character recognition model for reading the text itself, and an optional colorizing model for adding color to black and white pages. The README notes you can add or swap in new translation models yourself, as long as they use the GGUF file format, a common format for running language models efficiently on regular computers. A separate setup option through Hugging Face Spaces is mentioned but the README currently marks it as inactive. The README also lists some known problems and their fixes, mainly related to getting CUDA and its associated Python packages working correctly for GPU acceleration, along with a note that application logs can be viewed through the local web interface.