Trained on a mix of official arts, fan arts and anime screen scrapes. Style wise, it leans more heavily on the anime art style.

Please use v2 - it is the first Flux character LoRA that I feel have accomplished the quality level I wanted to see, hence coming out of the prototype state. It is stable, consistent and flexible, as demonstrated in the showcase pictures.

The main difference between v2 and v1 is the network dimension. I dropped it from ?? in v1 to 4 in v2, and the model stability increases dramatically. This seems to be the key for training an anime LoRA on Flux.