The lora is trained on the images based on the 1.5 checkpoints on here: https://civarchive.com/user/Sdive/models

1.5 is flawed so expect some images coming out weird.