Throwing pasta at the wall again to see what sticks. Played with the training settings too, and I really like how this one turned out. It has good strength to it, and matches the training images well.

In addition to the base style this LoRA should:

• Do plump / curvy / thicc / chubby girls well

• Do pubic hair and armpit hair well

• Easily do darker skin. I got tired of prompting 'very dark skin' and just getting medium brown. Melanin levels should go from tanned skin to dark skin to very dark skin. When using tanned skin put 'tanlines' in the negative prompt to avoid them.

• Dark skinned girls with a male should default to a light-skinned male. I also got tired of prompted darker skin for girls, only to add a partner and all of a sudden she's white and he takes the skin tone (even when specifically tagging 'dark-skinned female' the male would grab the skin tone more than half the time). Now I'd say 19 times out of 20 the girl will keep her skin tone and the guy will be light skinned. For a darker guy prompt 'dark-skinned male.'

If some of the scenes seem kind of random in this or my other LoRA that's because they are. I use a complex set of wildcards to write most of the prompts for initial showcases (and training data) to create a diverse and varied mix of images.

If you create something nice with this LoRA please share! I could stare at my own images any day. I like to see what other people create.

That said, someone on a previous LoRA mentioned issues getting the same quality. Every image here used WAI-NSFW 14, Euler A, and uses high res fix at 1.5x to up the size, detail, and smooth out artifacts from the initial gen.