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
Slight improvement of stability for 0.46.
No major new functionality added.
FAQ
Comments (9)
0.43 version seemed good, but where is the full fp16 or bf16 UNet? I would love to have it for LORA training.
I know the 0.3 is UNet but i want the 0.43 as a UNet checkpoint instead. Please make it available.
Unfortunately they are not available.
I trained them in fp8. And I have no UNets. It will be too much to publish everything.
Feel free to convert existing ones to it.
I recommend 0.46 for much improved rear views and stability.
Newbie here. Getting 'NoneType' object has no attribute 'tokenize'. Could you please provide a sample workflow using this checkpoint?
I don't know what that error means.
The preview images are generated in comfy, so you can use them as a sample workflow. Just drag them into Comfy :)
But anyway, the model should not be anything special. Well, depending on which version you are using exactly. Assuming you are using Comfy. But which exact model name are you using?
@6tZ Sorry for the delay replying. I'm using the latest 0.47 model, "fuxCapacityNSFWPornFluxBaseModel_v047DevFp8.safetensors". It was my first use of a checkpoint rather than a base. Turned out I needed to add a dualclip loader to the basic workflow I had downloaded, so I DO have it working.
This checkpoint seems to be heavily skewed towards bigger/fuller breasts. Nothing smaller like Dakota Tyler?
BTW what's the difference between the 160mb ae.sft you recommend vs the 320mb ae.safetensors that I already have?
Thanks for letting me know about dragging and dropping the images - I had no idea. That's neat!
EDIT: The image workflow threw up those errors as well, including a similar one for attribute 'clone' for the clip layer node.
@LMahesa Good to know it worked out!
And yeah, it seems like it's a big difference in how you load models.
Regarding ae.sft = I'm sure both are fine, I don't know any difference.
If it works it works.
Regarding breast size: I have found that it can do large/small, but it doesn't ALWAYS listen.
I have datasets to train this better, but Flux doesn't listen too well to instructions and captions.
@6tZ I'm actually aiming for the JAV / KPOP physiques - lithe, small breasts etc. Mixing with some asian loras works to a degree, but as you said Flux doesn't like to listen all the time haha.
@LMahesa Yeah!
I'll see about improving the quality of this, just by having more in the training data.
@LMahesa I’d still like to see a working node workflow. All I’ve been able to generate is a brown blank blob