NSFW model trained on a couple hundred titfuck/paizuri images. Uses 50% HassanBlend1.4 + 50% AnythingV3 as a start and merges well with models/blends such as SamDoesSexy, AnythingV3, SimpMaker, or HassanBlend to tweak the art style. Trained with split images so it also works reasonable well for landscape/portrait sizes.

The dataset used the following descriptors so they'll work best as positive/negative prompts:

photograph, illustration, pov, from above, from side, dutch angle, standing, kneeling, sitting, lying down, wet, shiny, pale, tanned, tattoo, lingerie, bikini, clothed, glasses, necklace, jewelry, looking at viewer, smiling, open mouth, cum

All example images were generated with either the standalone model or 50% merges with the other models mentioned above, the vae-ft-mse-840000-ema-pruned VAE, and this dynamic prompt (and ~20 generic negative prompts):

nsfw, uncensored, { photograph | illustration }, beautiful naked woman, { slim, athletic | }, { from above | from side | dutch angle | pov | closeup | wide angle | }, { lying down | standing | kneeling | bending over | }, (titfuck, paizuri, cock between breasts, penis), {2$$ cum | looking at viewer | open mouth | smiling | jewelry | clothed | wet, shiny }, { fantasy landscape, elf, queen, princess, mystic, pale | beach, outdoor, tanned | bathroom, spa, steam, fog | at night, elegant, city, midnight }, { black hair | brown hair | blonde hair }, { short hair | ponytail | long hair }

Also, yes, I know this is what Textual Inversion is generally meant for, but I simply did not manage to get satisfying results that way.