This is a model trained on a combination of artwork from DaveCheung/VanillaCookies/TehDave and artificial work from prior generations using loras trained on the same artist at about a 6:1 ratio in 211 images. The idea was to capture the facial features while preserving detailed backgrounds and high fidelity (to avoid a more sketch-like style). I also wanted to avoid the more desaturated colors of the source art. It worked out well.

This is my first lora. I needed it to maintain a set of facial features I've been leveraging across SD 1.5, XL, and Pony. It seems like nobody was going to tackle it for Illustrious, so I did it. Yay! Any donation of buzz is appreciated. Now that I've done one lora, I know I can do more, buzz permitting. There are some other things I'd like to do now that I know how.

There is one mild artifact to note. Due to strange circumstances, a few of the training images included mouse ears and maid outfits together, so the lora has cause maid outfits to be associated with mouse ears. You can defeat it with negatives (animal ears, mouse ears, bunny ears).