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
A gguf Q8_0 version of 0.49.
FAQ
Comments (33)
Some side-by-side same prompt would be the only way to appreciate these updates, I just bought the .46, why should I buy the 0.4X ?
Yeah, I should try to get a workflow going to follow the same prompts and structure for better comparisons.
But please don't buy every version. Get the free one!
Consider the "buying" (donating buzz) a way of supporting me if you like the results.
But to answer your question in this case:
0.49 is more stable. I get much more good images with it and fewer body horrors in general. It's more aesthetically pleasing as well. In terms of new functionality, not too much is added I'm afraid. It's a progress towards female + male simultaneous model, but it was a very solid model so I decided to share it.
Within forge GGUF Q8 v0.49 gives me the error "RuntimeError: mat1 and mat2 shapes cannot be multiplied"
Rarely happens and i've only seen this happening with 2 other Q8 checkpoints.
Any idea how to fix that?
No idea at all, sorry :(
You're saying it only happens sometimes for this model?
It sounds like a Forge issue then.
I'm staying away from Forge until it's more stable with things like this. Compared to ComfyUI it's not been too reliable. With LoRAs not always working, it's much harder to debug and figure out the actual cause of problems.
Which application did you use to train it? If i might ask. Is it Kohya?
@MrToon Yes, it's trained using the Kohya-scripts, via ComfyUI.
@6tZ Not happens with another model, its model issue.
@velanteg And what program were you using to generate with?
Same problem with Q4 v0.49 in forge.
v049DevQ4KSGguf.gguf gives error:
RuntimeError: mat1 and mat2 shapes cannot be multiplied (4032x64 and 256x768)
Please share your workflow. At least send some screenshots of what you're doing.
@6tZ Forge, Comfy too inconvenient to use.
@velanteg Okay, yeah I have no idea then. I don't know why forge doesn't accept it. I guess it reads models differently.
I have encountered a similar issue. My reason is that I adopted a controlnet model of a different type from the checkpoint model. It is recommended to replace it
hello good :D which of all these models can I use with 12 GB of vram??? thanks :D
on Q80Gguf.gguf with forge i got this "RuntimeError: mat1 and mat2 shapes cannot be multiplied" :(
Thanks for the notice!
It looks like the way I quantized the models isn't readable by Forge.
You could try downloading a full model, and re-quantizing it yourself with Forge.
I'll do it and re-upload at some point too.
What are the recommened settings to use this properly? Sorry Newbie
If you click on one of the images showcased by the model, you should see the generation settings on the details on the right side. At least whichever is picked up by the image saving node.
But generally: ~30 steps, a CFG of 3.5 and the Beta scheduler is what I use. Euler-A as sampler. But I think most things work fine.
Where them weiners at Bro ?
Not in this model unfortunately. This one is pretty set at the female genitalia. Any weiner-addition just gives you lots and lots of mussies :)
You'll have much better luck with the Fluxed Up and of course, Pulsar.
Hello! Does this model include VAE and Clip?
I don't think so. I think you have to load them separately. I do at least.
I could not get it work with the Flux basic workflow under ComfyUI, it says no Clip & VAE. Can you please provide a simple workflow for me as a beginner? Thanks a lot !
@liningbns859 Hi. Sure! In fact, each image posted by me here contains a workflow.
So if you download any of the images in the preview, and save it to your computer, you can then drag/drop this image into ComfyUI, and it should load the exact workflow.
You will need to link it to the correct models for yourself, based on your model paths etc. But that should be simple to do :)
@6tZ thanks,as a beginner this trick helps!
@6tZ why is this so complicated, i m unable to get it right, always missing nodes, i already restarted 10x comfy, switch many thing in version, and then you have 3 place to enter text, 2 place to put your resolution and the most important, 0 place to add your custom lora , i think i will just move one to an other checkpoint, and why are all pics with blurry background, can't it make normal ones, doesn't make any sense to me, you try to make something beautiful , if it's blurry it's useless, just put a white back on anything lol
@carlosurportable419 The workflow I use is not necessary.
It's complex because it helps me to keep things separated. It's not meant as a beginner friendly workflow. You'll have to find that elsewhere I'm afraid.
I use this workflow because it works for me and gives me the control and stability that I need.
Pics have blurry backgrounds with Flux because it's generally aesthetically pleasing and considered good looking for professional photography. There are LoRAs you can use to bring that down to get a wider depth of field.
Good luck! <3
One of the most promising engines, and compatible with loras. Thank you! I will follow updates
Thank you.
I have some newer versions of this but they are no good. It's a very unstable model. It's quite heavily tuned to specific types of images, and it doesn't like going outside of it as it is.