Pretty decent "portrait/modeling" support Lora. Used Dreamshaper, so you can see some measure of bias in the base model bleeding through. "On back" doesn't seem to work with the Lora though, and even with "NSFW, nude" in the negatives... well you can see my results. 😏
Might benefit from more variety in the training data to make it a touch more flexible depending on what the intended output potential is meant to be. Not a half-bad first draft though, I'd consider keeping it as hammer when I've got a nail to drive.