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 (24)
finally a proper flux pussy!
Server execution error: ERROR: clip input is invalid: None
If the clip is from a checkpoint loader node your checkpoint does not contain a valid clip or text encoder model.
I use krita ai, what vae do I need for this model?
Hi.
You need to load VAE and CLIP from other sources.
This goes for most flux models. At least the ones smaller in size.
I don't know Krita AI unfortunately. But you can find CLIP and VAE guides here:
You'll want both of these text encoders:
clip_l.safetensors
t5xxl_fp8_e4m3fn.safetensors or t5xxl_fp16.safetensors
And you can also use a clip_g.safetensor at the same time.
For VAE it's usually named ae.safetensors
I guess it's not (V)ariational.
Good luck with the setup!
@6tZ thanks for your help! My mistake was that I put fluxed up in the checkpoints directory instead of unet, now everything works.
@6tZ thanks you for me too !!!
Thx for the suggestion, I've now added this note to the model page!
"Breasts are fantastic and vaginas are often quite good." Kinda reminds me of my first month at university in the 1980s!
Excellent Work. Many Thanks.
A little update about the stripes over the images when upscaling :
There seems to be a solution : https://github.com/lllyasviel/stable-diffusion-webui-forge/issues/1712
The lora trained should be trained without the option "T5 attention mask".
I didn't tried yet, but it could be the problem with the recent versions of this checkpoint (assuming that you use a lora merged with the original Flux-Dev)
Thanks again for the advice! I'll see if my training setup has an option for that already!
I see it happening occasionally still, but it should be much less of a problem with the newer version.
I still see subtle remains of it, but much less noticable!
@6tZ : a little update :
I made a lot of test about the stripes on upscaled image, and here what I found :
- "T5 attention mask" option doesn't seems to change anything (at least for me).
- Over-trained LORA make stripes compared to "not" over-trained or under-trained LORA.
From my tests, if I use (for example) a LORA with 100 epochs at strength of 1.0 ==> stripes.
The same LORA but with 50 epochs at strength of 1.0 ==> no stripes.
It's difficult to be sure, because of course, at the same strength, a "less" trained LORA is less effective (I mean from my tests, a 50 epochs LORA at strength of 1.0 is approximately equal to a 100 epochs LORA at strength of 0.75 (this is an example and it's an approximation !))
But the important thing to consider is : I could have no stripes and good results (from my LORA) if I use "not" over-trained or under-trained LORA. Keep that in mind.
Cheers :)
PS : I don't know if you (6tZ) remember how you did your v1.0 compared to most recent one (and the number of epochs, and I believe the LORA rank could play too....), but maybe you could explain that with what I said...... (once again, assuming you are using LORA to make your checkpoints, which I could not be sure).
@zetofzone204 Amazing breakdown once again.
You really should write articles about this so more people can learn from your experiments.
Unfortunately I didn't keep any notes of what I used for the earlier versions.
But it was a mix of my FuxCapacity checkpoint and this one LoRA:
https://civitai.com/models/815791/fux-merge-tools-female-genitalia-v1-flux-nsfw-merging-model
This really started unlocking the capabilities for me, but it WAS absolutely overfit.
@6tZ
Thanks for the answer.
I've posted my comments on the initial link I put (forge Github).
Cheers.
Can't get it to work on Draw Things for MacOS.
Okay. Does that program support Flux?
@6tZ Yes i'm able to run flux. When I click "import custom model" after having selected the file, it doesn't do anything, just opens and closes some progress bar and nothing happens;
@nirvana169 I'm also using DrawThings. Currently it doesn't look like we're able to import GGUF files. Try importing the safetensors file
GGUF version is not working for me, I don't know if I'm doing something wrong.
Sorry to hear! What tool are you using? Forge or ComfyUI?
Which folder are you placing the model in?
I'm using Forge, the folder is C:\pinokio\api\stable-diffusion-webui-forge.git\app\models\Stable-diffusion
@Hyperion3601 Yeah I think that should be correct. This is similar to my Forge setup as well. The model works fine in there for me. What problems are you getting? Some error in the console, or some specific result? I got a black image once, but it was because Forge has reset my default sampler to [NOTHING] and it refuses to update the defaults.
Great Job :-) Does anyone know if you can use this checkpoint to train a LORA with AI-Toolkit? If so, what do you need to bear in mind?
Nice! Anyone got a decent workflow for Comfy? I'm new and can't seem to get it to work
hi - use my workflow - enjoy FLUX UNREAL - v1.0 | Flux Workflows | Civitai
@showbert7 great thanks v much