This LoRA aids in the generation of side-by-side clothed and nude images of a subject.
Note that you may need to modify the weight to prevent artifacts when working with models that are Illustrious-derived but which have been significantly fine-tuned away from base Illustrious. Generally lowering the weight will improve the output in these cases.
Description
FAQ
Comments (3)
Not sure about this one, but I regularly check Civarchive most recently deleted.
Seeing the OnOff LoRA I think it must be the best thing ever
Naturally I image stitch a...character and a blank image and mask the left side and try it on Nunchaku Flux.D
I must say the results were disappointing. Maybe this or the other ones will be better?
Back to tight masks, depth maps, and ancestral/~0.5 denoise or just the fucking Wan base model for me
Yeah, the nature of in-painting using diffusion models is such that the OnOff loras are not really useful for nudifying existing images. Which is good, because such tools are explicitly against the CivitAI terms of service.
Now if I can just convince the moderators of this, my OnOff loras can be republished.
@Invisidude Oh, that should be pretty easy. Just make workflows with a red-haired anime girl with InpaintModelConditioning with stitched image pairs:
Image 1 (send to pixels) source image - white image blank
Image 2 (convert to mask) black image blank - white image blank
It'll still make real women with brown hair. That'll prove it doesn't nudify characters.
Don't tell them about Flux UMO, ancestral sampling, depth maps, pretty much any video base model or anything else people actually use to nudify characters without making them gain pounds and lose inches though. That can be our dirty little secret.
You know, I speculate that barring refusals, Grok is probably better than anything else at that.