Illustrious, NoobAI-XL, Pony, and their variants already can create a great illustrated diaper, but they do their own thing with the style. Since I'm doing photoshoots for the Flux ABDL Brand loras, I figured it would be worth a shot at trying to use the same datasets for an illustrated model. Results are lackluster. That said, they do capture the essence of the diaper. It's splotches instead of proper prints, but there will be splotches in all the right places to remind you of your favorite ABDL brands.
'diaper' was the only relevant caption used in training. '(diaper:1.2)' seems to provide enough weight anywhere in the prompt.
For comparison, Base Illustrious and WAI-NSFW sample images are in posts below. Showcase images are NoobAI-XL-Epsilon.
All training images are photographs of sex dolls and mannequins in front of a green screen. Here are the captions I used in training. Feel free to use them in the negative if you are getting bad backgrounds: 'mannequin, photo-(medium), photo-background, green screen, simple background'. Results tend to be better without them though.
