The latest version has a bit less style, more realism. But with better anatomy.
Download these nodes from the manager to use the GGUF T5 model:
https://github.com/city96/ComfyUI-GGUF?tab=readme-ov-file
Check the files at the side:
200mb Clip L
2GB T5 GGUF
11GB FP8 checkpoint <<
Use any Flux VAE
I'm currently in the process of making Schnell more like DEV, Initial versions were merged with DEV, but now I've rolled back on that and focused on developing schnell solely. For 2 main reasons. 1 because it handles styles better than DEV, and 2 because the license is more approachable for all use cases. Hopefully this will become a decent base model to train upon.
With the addition of the Hyper 8 step lora, from ByteDance, areas where the model might lack coherence can be adjusted. (above hyper lora strength 0.1875)
Thanks to OldCelticSoul for converting FluxWoW2 into GGUF format!
https://huggingface.co/ND911
WoW_XL_7 was trained on an additional 20,000 images, and then merged with
Kohaku Epsilon
RealWoW4
Nightvisionv9
Mobius
FlashBackXL
Aura Render
It was also merged with speach1sdef178's latest works in which were trained an additional 20,000 or more images.
Great care was taken to make the model capable of a huge range of styles. Using the Mile High Styler and the default set of 1300 styles, I've meticulously tuned, trained, and merged around simple prompts in order to make sure that the model is as adherent to different concepts and requests as possible.
It is NOT an NSFW model, but it is also not "censored" either. If you want NSFW, choose the WoW_PD version instead.
WoW_XL_6 comes with an enhancer lora. Make use of both for best results in certain prompts! Especially with text.
SD3 Model removed for now until CIVITAI can clarify it's position legally.
--
The latest version of WoW_XL (5) is a collaboration between myself and https://civarchive.com/user/speach1sdef178
With https://civarchive.com/user/LawyerFrank also as a major contributor also.
Special thanks to both of them for doing such great work by themselves and also for giving me such great feedback and input over the last few months.
Frank in particular deserves extra thanks for helping financially support me while I was suffering from some health issues.
And speach1sdef178 for her unending support and amazingly positive attitude.
Also I'd like to thank General Awareness, for his assistance https://civarchive.com/user/general_awareness/models
Without his insight and experimentation in training none of this would be possible in the way that it has been.
And finally I'd also like to give mention to WarAnakin, Afroman4peace and Freek22 for their insights, feedback and also for setting their own standards of quality for me to work towards :D
WoW_PD, Version
This is a versatile branch of PD which looks towards a more realistic output, but with a focus on overall quality and flexibility.
This model is not good at artist styles, but is great with people, animal and fantastical prompts and is capable of a wide range of styles from CGI, realism, cartoons and anime to pretty much anything else you can think of that doesn't require an artists name to infer.
It's also capable of both SFW and NSFW content, with the composition itself remaining almost unchanged between the 2 styles of imagery, outside of the removal of clothing items and the repositioning of limbs in order to present the subject in a more erotic mannerism.
It does both male and female subjects well and can even place 2 separate subjects in the same scene together, participating in specific actions and wearing specific clothing items.
The main reason I'm creating this separately is because I would like to focus on a "WoW" style model here. With the aim being the most beautiful output possible, rather than necessarily focusing on any of the other details.
It's a work in progress, but I think you will enjoy using this version currently.
--
Version 5 of WoW_XL is out now!
WoW_XL Five
Massive improvements all round. Can now prompt for cartoon styles and realism styles simultaneously and achieve both in good quality.
I rolled back a bit on the rick and morty stuff and decided that I will work that into a dedicated cartoon model. And while this model is still superbly versatile it does realism just a bit better than before :D
As the name entails, this model is all about the wow factor. It's aimed at producing the most beautiful output you can possibly imagine.
V3 uses inpainting VAE and V3 Diff VAE uses SDXL fixed VAE
This model works on low steps, ~20 and at resolutions of 512x512 for those who want to save time.
Apparently my workflow is "too big" for Civitai, so I have to create some new images for the showcase later on.
https://www.buymeacoffee.com/tripleheadg
Models used to create this:
Kohaku Epsilon
RealWoW4
Nightvisionv9
Mobius
FlashBackXL
Aura Render
Crystal Clear 3D
Crystal Clear S,
RealVis4
JuggernautXL v10
HoloFormXL
Cataclysm - Artistic Pony realism
Trifecta XL
Pony Realism V2.1
Better Than Hentai XXXL
Kika-Real
Ratoskr
Tessilation v2.5
Kika
SoftPonyXL,
PonyDiffusion
SevenOf9PonyRealMix
AnimeGodXL v3
The Talos Project
XL6 Hephaistos
SimpleDiffusionXL
SpectrumBlendV2
7 of 9 V2 (yet to be released)
Samaritan 3D
RunDiffusionXL
NightVisionXL
Yamer's Merge Heaven
SDXL Base Model
Description
UNET ONLY
Thanks to OldCelticSoul for converting FluxWoW2 into GGUF format!
https://huggingface.co/ND911
Check out his hugging face here!
FAQ
Comments (21)
i appreicate your efforts
I don't see which quant the GGUF is? I'm assuming Q6 or Q7 based on the file size?
Hey there. Sorry, it is Q8!
@Triple_Headed_Monkey Cool, I only have 8gb but between SD Forge and my 32gb of sysram.. I'll give it a try! :)
@MysticDaedra I'm using it with the 2.1GB GGUF t5 model currently, that might help you somewhat :D
i wan to run gguf q8 on cpu , any help ?
In comfyui, try adding --cpu-only to the launch arguments
IMO that would be painfully slow. I would also think that fp8 would be faster than q8?
@NowhereManGo as my test ..it's perfectly slow .yes . but the gguf q8 is way faster than the fp8..,aight be memory thing ? but it faster than any fp.
how many steps at least needed for this q8 unet ?
8 steps :D
This is based on Schnell, so 8 steps for best result, but for quick testing of prompt 4 steps is good enough.
All your models are mere crap.Nothing works properly after hours of downloading, always something is missing.
I can quite definitely assure you that all of the models work perfectly. What we have here, my good friend is a case of "user error."
Now, if you would give me some kind of idea of what the problem is, I might be able to assist you.
Everything that you need to use the models is here aside from the T5 models. Which you should be able to choose which one you want to use yourself.
I just want to point out that not a single other person has made a complaint up till this point. Furthermore I have more than 15% of people who downloaded leaving a thumbs up.
For most projects, it is a measurement of success when one sees 10% or more of it's users return to leave positive engagement.
And so the likelihood that the problem is with the files that I provided is very slim in this case.
@Triple_Headed_Monkey here are a few small complaints, what is the tool support for nf4, can you share a ConfyUI workflow? I cannot find a loader for it (I'm sure I could write a python script to inference on it but that's not were all of the CivitAI users are at).
Also you have one model entry but under it you have SDLX versions and several different Flux format version combinations and even a Lora or two. It makes it hard for your model description to convey details specific to the individual models.
I did reference your workflow from your sample images, but still blocked with a security warning while installing GitHub - comfyanonymous/ComfyUI_bitsandbytes_NF4e I cannot for the life of me get it to lower the security settings despite updating the config.ini
@Triple_Headed_Monkey Successful models usually have a ton of image submissions. Maybe the usage instructions could be improved.
@DearLuck Can't disagree there. I've only just released this latest version of the model and most of my core users are on Discord. I won't lie this has all been relatively confusing for me also.
So I'm going to write up a proper tutorial on it.
I've Just been ill for a few months, so I've been a lot less able to do the small things.
@Triple_Headed_Monkey My bad, that was shit thing to say, I have looked at the recently uploaded thing and was tired too. In any case, can't wait for more tutorials, I hope we can settle onto one or two best ways to run things and not 16.
There is no need to be so rude. If you need help, just ask nicely. We are at the bleeding edge, so things are never smooth. One must read instruction carefully to get things to work.
These models are in GGUF format and are not regular safetensors, so you need to use ComfyUI with the appropriate GGUF custom loaders: city96/ComfyUI-GGUF: GGUF Quantization support for native ComfyUI models
You also need to install the VAE, CLIP, and T5 because they are not included. This is so that people don't have to download these rather large components again. If you don't want to do that, use this version, which seems to have everything included: https://civitai.com/models/147933?modelVersionId=745853
@Triple_Headed_Monkey Hope you are ok now, take it easy 😊👍
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.