I don't really know what you mean by "all the metadata" because I'm not entirely an expert in certain AI concepts.

I have the training params if that works for you, although I'm not sure if that's what you want.

Because the training was done right here in Civit, I don't have access to much more data beyond that. I also have the training data, but I'm not sure if I can upload it here (for copyright reasons), but if you want, I can send you the Mega link where I have it uploaded. If none of that works, I'm sorry, but I really don't know how to give you more data.

@magmafalcondec681 Oh, I understand now. Let me try to explain as best I can.

I use PixAI to generate the images I use; it's the site I normally use for image generation (because it doesn't have NSFW restrictions), so it's possible that not everything I explain can be done equally well in Civit (because, as I implied, I don't use Civit to generate images, so I don't know exactly how they work here).

First of all, the prompts I used are the same ones I attached in the images; I didn't include any more information because I only wanted them to serve as a reference of what it looks like, not to be recreated.

I use version 15 of the WAI-NSFW model, with a CFG Scale of 5, and euler a, for the sampling steps, I use between 20-30 (depends on the image), I use this negative prompt:

bad quality, worst quality, worst detail, veins:0.1, mole, moles, simple background, white background, (((poorly drawn feet)))

I also use face fix, and all the images were scaled from an original generation, the scaling doubled the size of the image 768×1032 → 1536×2064, I used denoising strength of 0.6, and the denoising steps at 50 (I didn't select an upscale sampler), I also recommend using LoRA at a strength of 0.7.

That's the best explanation I can give of how I generated the images attached to LoRA. I hope it helps!.

I don't usually scale images when I create them, but since my latest LoRAs I've tried to do so to make them look more professional.