MangaNinja

MangaNinja comes from a paper by ten researchers including Zhiheng Liu and Ka Leong Cheng, with the code published under ali-vilab on GitHub. The README describes it as a reference-based line art colorization method: you supply a colour reference and a line drawing, and the model maps the reference colours onto the line art. What sets it apart is manual control: you click matching points on the two images to tell the model where each colour should go. The interface works on 512×512 images, and the project includes a Gradio web interface that runs locally.

Who it's for:Manga and animation creators curious about AI colouring who have a GPU and are willing to set up the environment themselves. Readers who don't write code won't get much from it.

Before you start: You install it on your own computer: create a conda environment, download models such as Stable Diffusion 1.5, CLIP and ControlNet plus the MangaNinja weights, and you need a GPU. The online demo linked from the README is hosted on Hugging Face by a third party, fffiloni; when we checked, it showed a runtime error and couldn't be used, and any images you upload there go to that page. The local web interface listens on 0.0.0.0 by default, so other devices on the same network may be able to reach it; we suggest using it only on your own machine. The repository is licensed CC BY-NC 4.0, so non-commercial use only, with attribution; the other models it uses have their own licences. You need the rights to any characters in your reference images.

Based on the project's public documentation. Not tested by AI Inspiration World.