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
Quantized to Q4_K_S.
Unsure about the precision. The original model is fp8, but the gguf conversion process turns it into f16.
FAQ
Comments (32)
Training the CLIP_L text encoder helps.
Thanks for the tip!
I need to figure out how to do that.
@6tZ I wrote an article on when to and when not to.
https://civitai.com/articles/8108/training-flux1d-clipl-and-t5-benefits-and-risks
@AbstractPhila That's fine (I actually saw it already), but my concern was HOW :D
@sevenof9247 Huh? Waste money on what? Use your loras for what?
@sevenof9247 I mean, I'm giving away my $2000 lora, but it's all in research sake.
@AbstractPhila $2000 or buzz ... this shit is only started whit FLUX ... SDXL and PONY was all for free
@sevenof9247 Very well, thanks for your input.
Of course people should use your free lora! It's a no-brainer!
Nothing to see here, move along to sevenof8488747667657's loras.
Do check out the prompt understanding on 0.46:
https://civitai.com/images/32303325
Why no weiners Bro ?
Working on it! It's my goal for the next version. I want to make sure it's well made and flexible. Can't have flacid uncooked weiners!
@6tZ Bro thanks for puttin' in work on the starfish, now let's get some weiners up in this MF. Thanks !
@girlswithafros I still haven't been able to merge weiners into this. But here you go, double-weiner models:
https://civitai.com/models/884686/pulsar-male-gay-nsfw-base-model-flux
https://civitai.com/models/884734/fux-merge-tools-penis-v1-flux-nsfw-merging-model
Checkpoint and LoRA.
Checkpoint is better if you want just one.
LoRA is good if you want to merge yourself, or use on CivitAI.
<3 good anus
v.46 has seemingly placed all the vaginas on the pubic bone..
You get a burger pussy! And you get a burger pussy...! Everyone gets a burger pussy!!
And??? :-P
Very well done, my friend! I love that it doesn't COMPLETELY DISTORT the faces of my Loras... Nice touch... a little weird stuff here and there but EXTREMELY stable in general! I'm a fan and I've tried a BUNCH of crappy Loras and Checkpoints for Flux... I finally started making my own Loras so I could control the faces AND the bodies!! Wink, wink!!
Care to share your favorite Flux checkpoint, lora, workflow combo?
Get an error using this via Krita AI Diffusion: "Server execution error: local variable 'dtype_t5' referenced before assignment" (using version 0.43). Krita uses ComfyUI as a backend.
Which version did you use?
@6tZ 0.43-dev-fp8
@modelmixing Hmm, okay, yeah I have no idea. I've never tried using Krita via Comfy. Are you sure it supports Flux as it's inference model? Sorry if the model isn't working. It sounds like you need to load the t5 encoder, but I assume you have done this in the workflow already if you're running other flux workflows.
Some models have the encoders baked into the model, but I don't think mine does.
@6tZ I updated my Krita and Comfy server version and now I'm getting the same error another user is getting, about the none type object having no attribute tokenize. Anyways, just letting you know.
@modelmixing Could you share your workflow so I can test it please?
@6tZ You bet. I have Krita installed (latest version, 5.2.6), and I have added the Krita AI Diffusion plugin from their Github (https://github.com/Acly/krita-ai-diffusion). I have all the necessary Comfy and Pytorch installed, and all the other requirements (it walks you through it), using the "local managed server" option. I'm using the "Flux Euler Simple" option with 20 steps and 4.0 guidance, with your checkpoint loaded. If you need more info we can hop into a DM or chat somewhere.
@6tZ Here's the error report from Comfy via the browser:
# ComfyUI Error Report ## Error Details - **Node Type:** CLIPTextEncode - **Exception Type:** AttributeError - **Exception Message:** 'NoneType' object has no attribute 'tokenize' ## Stack Trace ``` File "C:\Users\******\AppData\Roaming\krita\ai_diffusion\server\ComfyUI\execution.py", line 323, in execute output_data, output_ui, has_subgraph = get_output_data(obj, input_data_all, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb) File "C:\Users\******\AppData\Roaming\krita\ai_diffusion\server\ComfyUI\execution.py", line 198, in get_output_data return_values = _map_node_over_list(obj, input_data_all, obj.FUNCTION, allow_interrupt=True, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb) File "C:\Users\******\AppData\Roaming\krita\ai_diffusion\server\ComfyUI\execution.py", line 169, in _map_node_over_list process_inputs(input_dict, i) File "C:\Users\******\AppData\Roaming\krita\ai_diffusion\server\ComfyUI\execution.py", line 158, in process_inputs results.append(getattr(obj, func)(**inputs)) File "C:\Users\******\AppData\Roaming\krita\ai_diffusion\server\ComfyUI\nodes.py", line 64, in encode tokens = clip.tokenize(text) ``` ## System Information - **ComfyUI Version:** unknown@6tZ Found this on Github: https://github.com/comfyanonymous/ComfyUI/issues/4868
@modelmixing Right, then it sounds like it's using the wrong setup/nodes for it.
It's possible that Krita doesn't support some types of models.
Flux model loading is very confusing, and there are too many ways to load models...
@6tZ Alright. Well, thanks for all the replies and for humoring me. I appreciate it just the same. But before I let you go, I thought I'd mention that the STOIQO Afrodite Flux model does work with Krita - maybe that could shed some insight. Cheers. https://civitai.com/models/207101/stoiqo-afrodite-or-flux-xl
@modelmixing Thanks for the tip!
It does tell me something, but I don't know what.
That model is 20gb. Maybe it includes T5 and / or clip? Or maybe it's a different float precision?
The model doesn't say what it is, and it's frustrating.
If it includes T5 encoder, and the workflow for Krita and Flux inpainting loads the encoder FROM the model, then it makes sense why this one isn't working. In that case you would need to re-bake this model with T5. Or update the workflow used by Krita to load the T5 from a Dual Clip Loader.
There are many ways that this could be wrong, and many ways to solve it. But very hard to do remotely without having exact information.
@6tZ Yeah and unfortunately it goes a little over my head, but it's alright. I'm sure things will settle out eventually.