What is this?
This LoRA is intended to patch an Illustrious model to produce better furry images from e621 tags. It should work with a variety of styles. I used the Illustrious v0.1 model as a base for training.
Usage:
I recommend using this LoRA with a strength of 0.8 to 1.0 for most cases. The best results seem to come from 20-50 steps with Euler Ancestral Karras/Beta (6-8 CFG), or DPM++ 2M Karras/Beta (4-8 CFG). Use e621 tags in prompt (with spaces instead of underscores) as well as quality tags e.g. "high quality, masterpiece, fox, male, anthro, solo, detailed background"
This LoRA can make models highly sensitive to the prompt so for best results use a large amount of tags to fully describe the image you'd like. Using booru tags like 1girl, 1boy etc also seems to produce more Japanese/kemono style images.
If you use ComfyUI, you can simply plug the lpp_e621 node from Lazy Pony Prompter directly into the prompt if you'd like (use the default EasyFluff style)