Instructions to use Comfy-Org/MiniMax-H3-Image with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Diffusion Single File
How to use Comfy-Org/MiniMax-H3-Image with Diffusion Single File:
# No code snippets available yet for this library. # To use this model, check the repository files and the library's documentation. # Want to help? PRs adding snippets are welcome at: # https://github.com/huggingface/huggingface.js
- Notebooks
- Google Colab
- Kaggle
Its fast but image quality is bad - washed off
Since it just came out i wont blame it maybe a few days of community work on it will get us to a much better quality level
Maybe you can try and increasing the resolution. From the official model card: "The effective working range of MiniMax-H3-Image is about 4 MP to 20 MP, and it generally gives the best results at around 16 MP"
Since it just came out i wont blame it maybe a few days of community work on it will get us to a much better quality level
can you confirm which rosolution you used?
even at 16 megapixels, it's lackluster so far. unless the community comes up with some tricks, I was highly disappointed since their video model is so capable. to call this a SOTA is a far fetched glazing.
um its actually not bad if used properly and after a lot of experimentation i have started seeing some good results although i did add a realism lora ( i wasnt sure if it will accept it since the lora was for the video model but it seems to be working ) - and unlike minimax i stopped asking it for realistic skin since it already produces realistic results and asking for it over does it
i generate at 16MP then downscale it to whatever size i want.