Welcome to The Fux!
The goal of this model is to unlock full NSFW capabilities in Flux, one step at a time, starting with a photorealistic female focus. Enjoy the show!
Recommended Settings
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...
Description
FAQ
Comments (28)
How does Flux do porn? You need apply loras one by one?
@justinchandler892 Not with this checkpoint. You need to use this model for it, or others that are NSFW-capable. Unfortunately, Civit does not allow generating with custom Flux models onsite.
can i use it for fooocus ?
Not the original Fooocus, that doesn't support FLUX but Ruined Fooocus does.
what about inpainting? will it work? I've installed ComfyUI-Flux-Inpainting.
I haven't tried actually. But it should work out of the box. Let me know how it goes!
Just thought I'd share my experience with 2.0. Using character loras I got very blurry faces with DPM++2M and Beta, but much better results with Euler Ancestral and Simple. Pretty damn good checkpoint. Thanks!
However despite my praise, this model does have a strange habit of creating intersex individuals, lots of hermaphrodites despite careful prompting!
@Hazman Yeah, it's not intentional, but a flaw of the learning. More training is needed to separate the concepts of which genitals belong to which person. It's something that even happens with SDXL/PDXL/ILXL sometimes, but it's much more prevalent here as this has received much less training, or, brainwashing of the models, if you will.
@6tZ of course not! I've only done a few loras myself, can't imagine how difficult it must be to get a good checkpoint. Keep at it my friend.
Typically DPMPP_2M works with KERRAS and DPMPP_2M_SDE (and *_GPU) with KERRAS and EXPONENTIAL and SGM_UNIFORM.
Hope this helps. It's my experience that some combos of sampler/scheduler always give crap results, whatever the model you feed them. Some combos just work. Some seem awful and make yellw-green-artifacts. It's a lot of fiddling about sometimes. Find a combo that works and you're okay.
Additionally make sure your VAE doesn't have the well-known glitch of making black-rectangles when using FP16. Search for and download a 'FIXED' version of the VAE. There are a few out there.
The loras trained with flux.1D doesn't works
Pruned. Their data-blocks won't match up so that's to be expected.
Hello,
You mentioned nothing is baked in.
I just wanted to clarify what is optimally to be used, I am currently using ae.safetensors along with T5xxl_fp16.safetensors and clip_l.safetensors.
Is this ideal or do you recommend something else, this being for Forge.
Thanks for the model.
You're good to go!
You are using the same things as I am. So this is what I recommend.
I guess, I would still recommend ComfyUI over Forge for stability, but as long as you have it working it's all good.
Can I ask how you are able to use the model with T5xxl_fp16.safetensors and clip_l.safetensors in Forge? I cant find any UI to add the clips etc. I sense I am misunderstanding something here.
when trying to run a generation with the checkpoint I get a "AssertionError: You do not have CLIP state dict!" error... I do use ComfyUI also, but I use Forge for certain things, and I would love to get this working there. Cheers
Thank you for your hard work. Can you please explain the the difference between Fluxed Up and this one?
Cheers and thanks! They have completely different base trainings and merging. Fux Capacity was based on a much more intrusive initial LoRA training and merging, so it produced better looking images, but more failures / body-horror as well. Fluxed Up was less extreme, and gave more overall pleasing results, but less hardcore if you will.
Now both are more relaxed and functioning.
In general, the Fluxed Up one is now used with more experimental training and techniques.
I hope that helped a bit.
In the end, it doesn't really matter, use the one which has images that you like the most.
Or try them both?
@6tZ You explained pretty well. I guess I will stick with Fluxed Up since hardcore is my thing and more pleasing images are what I after for. Thank you for your help!
Sorry, I'm a noob. Does it work with FooocusPlus? If so, how to use it?
I have no idea, sorry.
Reading on the project page for Fooocus, it says it's not supported. But I don't know Fooocus plus. Assuming Flux models are not supported. You may want to look into ComfyUI for this.
@6tZ just had a look at the github for Fplus, it does support flux, but lots of people are saying its currently very buggy. I would recommend SwarmUI which runs this and other flux models well.
@Hazman Thanks for the suggestion! If someone asks again I'll suggest it :)
RuinedFooocus runs Flux models really well. If you want to use Fooocus.
Flux models are on the generator now, so this model is accessible for this current week. Give it a spin!
V3 when :)??
@Rudycont123 On my way there! Releasing 2.1 now, but started tinkering with a new version as well ^^
We'll see how much different it is. I save main numbers for larger changes.
@6tZ Just saw the Fluxed up one as well!! LETS GOOO!! already got both