Trained on 150 works by Rice-Chan. https://twitter.com/RiceChanArt

I didn't realize until after training how crusty a lot of the dataset was. Highly recommend to use "jpeg artifacts" as a negative prompt, which seems to counteract it.