Lora to generate beautiful sexy Ladyboys mostly, but not exclusively, of Asian ethnicity.
V1 - Original dataset made of about 50% Full Body shots and 50% Genitals Close-up shots.
V2 - Almost the same training dataset as V1 minus a few lower quality images and cleaned/removed a few watermarks. Also this time, only 4000 steps (instead of 5000) which I believe looks cleaner and more flexible.
V3 - different set of images and composed exclusively of Full Body shots. The output feels a bit more generalized and have a different diversity scale, with a bit more Caucasian and Brazilian TS body types in the data (better chance of generating non-Asian without prompting).
Overall: no version is inherently better than the other, they all have their strengths and weaknesses. Personally I use them all randomly for added diversity.
Suggestions: combine with a Breast Size slider lora and/or a specific phenotype (Asian/Indian/Hispanic/Caucasian) lora for a much wider diversity. Can also combine with a penis size lora or a uncut/foreskin lora for better control on the genitals.
Strength greater than 1 will tend to generate a somewhat more average/similar face to the input dataset. Lower the strength in [0.7 - 0.9] range for more varied facial diversity.
Description
FAQ
Comments (2)
Was it trained on faces in the full body shots?
Not exactly sure what you mean. Full body shots by definition has faces in them. It is meant to try to map body types but some faces learning vectors is somewhat inevitable. That can be only tweaked via strength at this point.