Never watched the show, but here it is
Lora seems to be fixated on the default outfit. Reducing weight to 0.75-0.8, and describing the desired outfit before the trigger seems to help.
Dataset had no nsfw art, but illustrious and pony checkpoints are creative enough to make it happen
Description
FAQ
Comments (1)
"Dataset had no nsfw art, but illustrious and pony checkpoints are creative enough to make it happen"
Everything I've seen and researched suggests the idea of character LoRAs needing to be made NSFW capable is (almost?) purely an SD1.5 leftover. 1.5 LoRAs essentially considered a character's outfit an immutable part of them, unless they were trained on pics with multiple outfits or nothing at all (nothing at all...), and try as you might pieces of their original outfit kept sneaking in no matter what you did with the prompt. (styles likely still want NSFW data included if it exists to get a more accurate imitation of the kind of content content many users want, but it's not needed, but that's beyond the scope)


