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 (34)
The constant obsession of all nsfw checkpoints for shaving is really annoying. Where are actual natural woman? Why can't we have real normal nude woman?
Thanks for your feedback.
I'll add it to my list of things to see if I can balance out.
As it is, it's hard to get enough prompt understanding into a checkpoint for NSFW concepts. It barely listens as it is =/
I have created a lora specifically to add pubic hair, it works well with your checkpoint, but that means one more lora and the more I add loras to get somewhere each lora lose their integrity. I haven't figured out how to use several loras at full force when one is acting on a face and the other on the body... hence why I was looking for a natural nsfw checkpoint
@sparkleboy599 Right, I get you. I can try to merge it into a future version, but the question is how controllable it is. It has to add the option for it, not force it. I'll give it a go.
EDIT: Ah nvm, it's not a public LoRA. If you post it, let me know :)
@6tZ yeah sorry I didn't post anything yet. But I am interested to learn: how to you actually merge it? I have noticed that my forge UI has a great tool called super merge, but It has a ton of parameters and I am not sure how to use it, can't find any decent tutorial...
@sparkleboy599 I use ComfyUI for my merging. I load the models, and use some merge nodes, or just load the LoRAs on top and re-save it with a "Save Checkpoint"-node. You can do a lot in there.
maybe because pubic hair is unpleasant
@sparkleboy599 Didn't you get the memo? Normal is not allowed anymore. 😂
This is still a good model though!
Really wonder why flux is so censored
Is it possible to have a Q6_K GGUF ? It is the best compromise for cards with 16GB VRAM
It's certainly possible. It's just a bit of work to do for not enough value as very few seem to care about it.
You can definitely convert the fp8 model yourself to this size though!
@6tZ I thought only fp16 models were convertible
@Brewce I make my conversions with the fp8 model, it seems to work just fine. I never save the fp16 models, fp8 is good enough ^^
@6tZ OK, I guess it's the same for all the models found here, so i'll start to make my own GGUF from now !
Thank you for your answers.
The model 22 is biased towards nude women too much compared to 21. Even when I write "red dress", it still shows a naked woman or a naked bottom.
Yeah, the entire model is overfit on naked data. You can always merge it back a bit, but you'll lose some of the details of the nudity, and you'll get more body horror images of limbs coming out of places.
it has a potential to become something better in the future, yet, its a way too far from being 'polished', too many artifacts, quality more 'SDXL' kind, despite some minor recognizable achievements in this model it does not feel to have the quality and looks not realistic enough in the way Flux promise to be
Agreed, it's also not yet controllable. I think further proper finetuning is needed compared to merging LoRAs.
To people, who cry for hardcore! You can sleep well, it wont happen in the near future.
The problem is not flux, but t5. T5 made by google and censored by google. If tech guys manage to crack, train, whatever that shit, there will be porn models.
There is an exception though. Pony V7 actually managed to make hardcore, and very likely it uses t5, because the same resource hog, like Flux.
So, there is hope for you.
А разве уже вышла Pony V7?
@frostik1234887 No.
away from hardcore output , did google t5 has really limitations other than that ? might be as always that text encoders is the weakest point in text_to_ANY models ?
@amazingbeauty I have no idea personally.
Stupid question, but does it run w character Loras trained on Flux Dev? Caus I had been testing some, and results looked very strange when compared to just using the plain model
It seems to work well, at least when using it in ComfyUI. When using LoRAs in A1111/Forge I get really strange result as well!
@6tZ ComfyUI user here :-) thnx for reply
Would anyone have a workflow using LoRAs for this checkpoint please in ComfyUI they would be happy to share please?
@Yossii I can share mine, but not sure actually how to share here on this platform
@dwmuell574 I’ve PMed you :)
@Yossii You should be able to use any "LoRA Loader" just as usual. It works fine for me.
sorry for stupid question
but does anyone can tell me how can i use this model on forge ui ?
You download the model, then you put it in the folder webui/models/stable-difusion, you should also download the VAE, clips and text encoders and put it in the right folders, but at this point I'm realizing that is problably better if you just watch a youtube video about how to run Flux in Forge.
thanks a lot for your helpful info
i did your advice and now still working :)
thank you man
@Arshavin_arfrd Really? Awesome, have fun!