LoRA model Trained on the three anime movies of Madoka Magica

This is an update on a previously trained model that was trained on the tv anime, also there was an attempt to add back in some flexibility by curating the dataset diffrently. Please give me feedback if it did add flexibility or the model is better or worse than the previous version.

This is a Style Lora so if you want to get the characters back use tags most associated with the characters to steer the model towards then

examples:

Madoka: 1girl, short_hair,  twintails,   hair_ribbon,  pink_hair,   pink_eyes,  short_twintails

Mami: 1girl, solo, large_breasts, blonde_hair, hair_ornament, twintails, yellow_eyes, drill_hair, twin_drills

Kyoko: 1girl, solo, long_hair, red_eyes, bow, ponytail, hair_bow, red_hair

Sayaka: 1girl, solo, smile, short_hair, blue_eyes, blue_hair

Homura: 1girl, solo, long_hair, black_hair, purple_eyes, hairband,

ect..

if stuck on how to steer the characters inputting an example image into auto tagger will help.

This model was trained with labels constructed using smiling wolfs wd-v1-4-swinv2-tagger-v2 at .25 threshold. Tags outputted from that are likely to match the training data.

Suggested base model to use for the lora is 7th animeC