Turbo LoRA distillation of Kirazuri (Anima) v4.0

Important: This LoRA is not expected to work with other base models.

Generation Settings:

4 steps, CFG 1

sampler/scheduler:

res_2s + bong_tangent

res_multistep + simple

dpmpp_2m + sgm_uniform

Some artifacting is noticeable with samplers that work better at higher step counts i.e. euler

Tried to address this a somewhat in version 2.0, and these scheduler combinations also appear work much better at the low 4 step count.

Otherwise, please refer to recommendations on the Kirazuri (Anima) base model card.

Training details:

Created with DP-DMD diversity-preserved few-step distillation.

Thanks to sorryhyun for the trainer: https://github.com/sorryhyun/anima_lora

Dataset: https://civarchive.com/models/929497/aesthetic-quality-modifiers-masterpiece