This lora is trained on a dataset of regular feminine body with realistic large erect penis under their clothes. Clothes in dataset included mostly panties, shorts, briefs, leggings and some jeans. It is trained on close-ups, so do include a little something about the face of the subject like « looking at the camera ». Strength seems to mostly affect reliability on these, not size. To adjust size, it seems that the use of the trigger words used in captions are the best solutions. They should be optional, without trigger words it should still give you somehthing, but smaller.