Cruddy first version of KF1 lora. Generation tags are "kf1, kf1 style" and I suggest adding "shading" to the negative prompts. Trained on anything v3.

KF1/Irodori Style LORA (cruddy second version)

Dataset is deepdanbooru tagged screencaps of season 1. Model seemed to stop working well after the first checkpoint (afterwards would just generate noise) and the dataset was like 2,000 or so images. Also ignore that the astolfo example was generated with 1girl i forgot to change that tag

Files with SHA256: 5b62af1e8f4b7c11827ebb1b5fd0d906620935e4b90f8f09bce86d7eef251845