First and foremost thank you to @konyconi and this amazing LoRA guide. I was able to take the config settings and apply them to a Kohya LoRA Dreambooth collab as I need a much better GPU. The captioning tips were key I believe and I think there is still room for improvement on my end.

This LoRA was trained with 50 images of women with copious amounts of ????? on their face. In my testing it will cover the entire face, but struggles with other body parts, hence the name.

Suggested LoRA Weight:

Anywhere between 0.4-1

0.4 tends to be a more conservative amount, singular load range

0.6 or more gets you into the ??????? range

When paired with other adult themed models/LoRAs you may want to bring the weight down. Works fairly well with every modal I tested.

Suggested prompt includes:

cohf, ????? woman face covered in semen

Inpainting

Does a decent job of inpainting, leave restore faces off as it removes everything this LoRA adds.

Future Improvements:

• more training images, would like to get to the 100+ range

• I would like to cover other parts of the body, but that may require a name change

• simpler captions when training, "face covered" in prompts will occasionally get you an image of someone wearing a mask