Lora based on DenzelTIP's work.

V1 - MLP training data only, it looks more pastel and flat compared to V2, still useful for MLP.

V2 - Includes lots of furry training data, should be more flexible and handle other anthro characters better.