original content based on model definition by @Defozo. this model will hopefully allow users to experience the fp16 model content on lower hardware spec'd computers.
This is a merged model of a base model flux1-dev fp16 and NSFW MASTER FLUX LoRA : https://civarchive.com/models/667086?modelVersionId=746602
Version 1.2 has added PussyDiffusion - Flux Lora merged in to improve vagina detail:
https://civarchive.com/models/983498/pussydiffusion-flux
Version 1.2 hyper 8step models have the content of version 1.2 plus ByteDance Hyper-FLUX Acceleration LoRA has been merged in allowing images to be generated in 8 (sometimes 9 or 10) steps:
https://civarchive.com/models/691446?modelVersionId=774008
NOTE: When ByteDance Hyper-FLUX Acceleration LoRA is merged into a model and the model is converted into fp8 precision, the conversion process screws-up the hyper 8step acceleration blocks from the lora. This causes the model to behave like a standard (non-hyper) fp8 model. The good news is it can still be included in the image prompt and still get the benefit of hyper 8step with fp8 model.
If you like this content, try a version of my other model series:
https://civarchive.com/models/967270?modelVersionId=1171422
To convert and merge, I used scripts from:
https://github.com/kohya-ss/sd-scripts/
To convert to gguf images, I used scripts from:
https://github.com/Zuntan03/EasyForge/
Using --ratios 0.8 in the merging command for version 1.0 and --ratios 0.65 for version 1.1
requires ae, clip_l and (t5xxl_fp8_e4m3fn or t5xxl_fp16) vae/text encoders
creates fairly good nsfw content. no key words. tested with base flux image generation settings.
NOTE: if you're not interested in clothing, use the version 1.0 model. it does a much better job with female anatomy. version 1.1 sacrifices quality to work better with clothed women.
Description
change nfsw master flux lora merge ratio from 0.8 to 0.65 to prevent nipple and vagina from bleeding through clothing.
FAQ
Comments (32)
Great model thank you for sharing
glad you like it.
I got "AttributeError: 'NoneType' object has no attribute 'sd_checkpoint_info'" message when i run these version and the previous version too, any help plz?
are you running on forge webui? model needs ae, clip_l and t5xxl_fp8_e4m3fn in vae/text encoder box and uses euler sampling method and simple or beta schedule type.
@tedbiv Ya, I did that, but got another error: RuntimeError: stack expects each tensor to be equal size, but got [256, 4096] at entry 0 and [293, 4096] at entry 4 I didn't get what the issue? and yes I'm using forge webui
@Starfish88 what size images are you trying to create? most of my testing was 1024x1024, 1024x1280, 768x1344, 896x1152. i remember getting that tensor error once... i don't remember what it was. make sure diffusion in low bits is set to automatic and cfg scale is set to 1
are you able to run flux-dev-fp8?
@tedbiv Oh it works fine now! I think it was forge webui error, all the setting was right. Thnx bro, great work.
@Starfish88 no problem. glad it worked for you. enjoy... :)
ERROR:RuntimeError: Error(s) in loading state_dict for IntegratedAutoencoderKL: size mismatch for encoder.conv_out.weight: copying a param with shape torch.Size([32, 512, 3, 3]) from checkpoint, the shape in current model is torch.Size([8, 512, 3, 3]). size mismatch for encoder.conv_out.bias: copying a param with shape torch.Size([32]) from checkpoint, the shape in current model is torch.Size([8]). size mismatch for decoder.conv_in.weight: copying a param with shape torch.Size([512, 16, 3, 3]) from checkpoint, the shape in current model is torch.Size([512, 4, 3, 3]).
are you running on forge webui? did you use ae, clip_l and t5xxl_fp8_e4m3fn in vae/text encoder?
@tedbiv I've try to use it on Forge and ComfyUI,but bouth them don't work,only model 1.0 is working.
@tedbiv yes.ae, clip_l and t5xxl_fp8_e4m3fn,I'm all in.
@daniel691020314 that's weird... the only difference in generation was lora ratio? although version 1.0 is better for nudes :) are you using any loras or just this model?
@tedbiv just only it,cause I love your model ! it's very beautifu l!!
@daniel691020314 thanks. it looks like it's pulling something with fp32 weight. version 1.0 and 1.1 are both fp8. i'll have to do some digging. i remember getting a similar error when using fp16 text encoder...
where can i find the AE file?
@var9324695s instructions and links here https://github.com/lllyasviel/stable-diffusion-webui-forge/discussions/1050
Could you also create a GGUF version?
i'm in the process of getting llama.cpp stuff up and running. currently the convert_hf_to_gguf.py script errors out saying it cant find torch module, but, pip list shows torch is installed... so i'm debugging. when i get the scripts running i'll create a q8 gguf version.
i haven't forgotten you... i've got the llama.cpp stuff built and running. now i'm running into issues with convert script not finding an associated config.json file for the created models. so i'm off trying to figure out how to create them... stay tuned. :)
this msg keeps coming up on Forge
AssertionError: You do not have VAE state dict!
even I have the 't5xxl_fp8_e4m3fn.safetensors'
plz help me~
requires ae, clip_l and t5xxl_fp8_e4m3fn vae/text encoders
The best NSFW flux model there is at the moment IMHO.... :-)
thanks. if you're not using clothing use v1.0 it does better/more consistent nipples than v1.1. i'm working on trying to create a q08 gguf version, but i'm having issues trying to create config.json files for the base models... :(
@tedbiv Thanks for the info, Yeah a q8 version would be amazing!
is there Quantized GGUF version?
not yet. i'm working on it. i'm having problems with conversion scripts. they need config.json files for the model and i don't know how to create/what content is needed. it's a work in progress, i want to create a q08 gguf version...
so i've created a q08 gguf version, but it makes crappy images when used with loras. is this normal behavior for gguf models?
@tedbiv in ForgeUI you have to select on the top the Diffusion in Low Bits Automatic Fp16 and it should work? i imagine comfyUI has something similar in some node
@elguachiiii i'll try that
@elguachiiii yay! thank you, that worked... too many moving parts for my old brain.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.