| Hi LocalLLaMA, I created a post a few weeks ago, but this time this project has become more reliable and easier to use. This is a manga translator that can also be used to translate any image. It uses a combination of object detection, visual LLM-based OCR, layout analysis, and fine-tuned inpainting models. I believe it is the most performant and easy-to-use pipeline for manga translation. For the LLM part, I have integrated llama.cpp into this application; it supports the Gemma 4 family and the Qwen3.5 family, and also includes uncensored and fine-tuned models. It also supports OpenAPI-compatible API, so you can use LM Studio or OpenRouter, etc. I think the demo video explains the workflow a lot, basiclly you just click a button and it will run the pipeline for you. You can also proofread and edit the result, changing the font, size, color, etc. It's a mini Photoshop editor. For who may have interest on this, it's fully open-source: https://github.com/mayocream/koharu [link] [comments] |