Amphibia Diffusion

Originally posted to HuggingFace by Theodora527

fter almost 3 months of work, ever since the first finetune tools were made accessible, I can finally release on of my first public models. I been also working on many other models based on various cartoons, and hopefully I can release them soon too.

The human generation is most problematic, SD really struggles to generate cartoon faces, as few wrong strokes completely messes up the face. I tried my best to fix it, but after many attempts it seems the only way to 'fix' it is by not providing varied face expressions. However most mistakes should be easy to fix manually, especially missing eye pupils.

Currently the various amphibians like frogs, toads, newts are NOT part of the model. I have dataset almost ready for them, I will train a new better model once I have more time. The big dream is also to train major and other characters from the show. The backgrounds from human realm are also not trained, so results will be bad if you generate it, and I might have to train in on separate token as the art styles will conflict each other.