>>10733
I'm using a lot of steps to get the final output
1) use model like flux with NSFW lora, to generate "man is sitting holding a ball on his penis"
2) use a context model like flux.kontext, "change ball to wear a face mask"
3) use Stable diffusion model, with regional prompting, setting the mask layer over the face mask, and setting the prompt for that region to be "1girl, ayaswan, character, head_only"
sometimes, I also have to paint in bits here and do a lot of img2img to get the final output. its a look of trial and error.