A nude/NSFW capable model that focuses on images with a female primary subject.
It is heavily biased towards nude images.
Recommended Settings
Sampler is dpmpp_2m (DPM++ 2M) and the scheduler is beta
No VAE or CLIP is baked in. Use separate sources for those.
Each sample/preview image contains the used workflow. Here's a quick article with simpler more beginner-friendly workflow. This is a recommended starting point.
FP8 Download
For some reason, when you upload multiple formats (BF16 + FP8) on one model, which they now support, they hide the FP8 under the BF16 model group in the side menu...
Early Access is enabled to support the development of new versions and models.
Description
FAQ
Comments (15)
Your model seems promising. What sampler, loras etc do you use?
These images are generated purely by this checkpoint. There's no LoRAs loaded on top.
Sampler is dpmpp_2m, and scheduler is beta.
It should be in the metadata, but since Civit doesn't load everything from Comfy, it's not all there I guess.
It's likely in the image metadata though, if you pull one into comfy you'll get the entire workflow.
@6tZ Ok thx, maybe you should put some infos in the description. It can be more appealing for the potential users
@wayenote4205557 Good idea!
Can you please include a Q_8 gguf or nf4 quant? fp8 is old news
How is it old news exactly?
@psspsspsspssspss I don't know how to do nf4, but I can try a Q8_0.
And please tell me why Q8 gguf is better than fp8?
It's larger and worse in quality.
With nf4, at least it can be run on smaller hardware, at a quality loss.
Q4 would be good
@superuser111 I can make a Q4, but not nf4. Don't have the tools for it =(
@6tZ Sorry for the slow reply, q_8/quantized models encode an fp16 model to higher accuracy than a normal fp8 model. A q_8 gguf is much closer to the original fp16 than fp8 is.
@psspsspsspssspss Thanks for the info! But my model isn't available in fp16 to begin with, so the data wouldn't be there then, would it?
@6tZ Yeah, that's unfortunate. If you retrain on dev fp16, we'll get even higher quality output if you then upload a q_8. Don't sweat it with this version, but consider it for the future. Even at fp8 I'm getting pretty good results :)
From my own experience, FP8 is better quality and faster on low VRAM systems than Q8 GGUF. NF4 is obsolete already in favor of GGUF.
@Learning2023 I think so as well.
But if it's a max vram optimization though, it does make sense to support it.