Note: This needs to be used with a de-censoring LoRA. It does not work by itself.

I tested with this one: Krea2 TextFusion Refusal-Reduction LoRA

Trained using AI toolkit on 172 images of flaccid uncut penises with foreskin overhang.

I did not caption the dataset except for the trigger "penis". But it still works to specify small/large, hairy/smooth, etc.