Style based on Akasaki's Bunny Girl MV. I extracted some key frames from the video. Since there are basically just two characters in the video, I trained a base model to help me draw new versions of the same frames, introducing character variety. This model was trained with the manually-extended dataset. As usual, I trained on WAI v12.0 as checkpoint.

The trigger word is akasaki_bunny_girl_style. The LORA weight works fine at 1.0. You may reduce the trigger word's weight for specific characters. It has worked fine between 0.8 and 1.0 for me; take a look at the sample images.