First attempt at training a Lora. I had very few images to work with, but overall it does fairly well at capturing the character. In some models and prompts, it will lose either the hair or the eyes and you'll have to add them back. I tend to use "aqua_eyes" and "purple_hair."

I've had decent results around 0.7, but the Lora does seem to mangle hands a bit so going lower can be helpful. Depending on the model I still get Amid with values as low as 0.5 and the hands do a bit better.

NSFW tag because nudes were used in the training.