A collection of GGUF for Illustrious models just to help low memory people and do some experiences:
Animagine - https://civarchive.com/models/260267
Bismuth Illustrious Mix - https://civarchive.com/models/1359028
Bubble hentai - https://civarchive.com/models/1164420
CherryMIX_XL - https://civarchive.com/models/1079112
CyberRealistic - https://civarchive.com/models/1125067
GoldenHour - https://civarchive.com/models/2433139
GSBeyondReality - https://civarchive.com/models/2455942
Hassaku - https://civarchive.com/models/140272
Holly Mix - https://civarchive.com/models/959490
iLustMix - https://civarchive.com/models/1110783
Nova Anime - https://civarchive.com/models/376130
Nova Furry - https://civarchive.com/models/503815
Nova Orange - https://civarchive.com/models/967405
Nova Reality DMD2 - https://civarchive.com/models/453428?modelVersionId=2439777
One Obsession - https://civarchive.com/models/1318945
One obsession Branch - https://civarchive.com/models/1368727
PerfectDeliberate - https://civarchive.com/models/24350
Poly-Goth Gloss - https://civarchive.com/models/2506452
Prefect illustrious - https://civarchive.com/models/1224788
RagIllustriousMix - https://civarchive.com/models/1491577
RealPhotoIL - https://civarchive.com/models/2041366
Ri-Mix Omega - https://civarchive.com/models/996495
SmoothMix - https://civarchive.com/models/1695253
SmoothMix Ultimate - https://civarchive.com/models/2311905
TovaMix - https://civarchive.com/models/2284900
Unholy Desire Mix - https://civarchive.com/models/1307857
WAI-illustrious-SDXL - https://civarchive.com/models/827184
The rest of the GGUFs versions are in my huggingface - https://huggingface.co/BigDannyPt/Illustrious-GGUF-Collection, not going to upload since it takes a lot for it in CivitAI.
The hugginface link also includes the clips and vae for each model.
What I noticed is that Q6 or lower, will increase the s/it to double, not sure why...
If you have a specific model that you want the conversion, you can ask that I'll try to do it
Description
FAQ
Comments (10)
I am completely ignorant on the AI scene, I have trial and errored comfyui into producing images by slotting in checkpoints and loras but have no understanding of the underlying mechanics or what can actually be done beyond generation.
Working under that frame, what is a GGUF model and how do I use it?
GGUF is a quantization of the model, consider it as a ZIP from the original file. And depending from the codex that was use it will compress more or less, but it will degrade as it compresses more.
I can only help you with ComfyUI, you would have to use the ComfyUI-GGUF node to be able to load them. Add the GGUF into the /models7unet folder and then that node will be able to read it.
From a default template/workflow, you only need to replace the model loader for the gguf loader.
Hey appreciate the effort, checked few models and its usually %20-30 faster + less VRAM invensive as well so pretty good I would say.
One thing, do I need a seperate VAE for every model? Isn't there a general VAE for all? I was normally using this for all .safetensor models for example (I am beginner so it may be a noob question)
https://huggingface.co/madebyollin/sdxl-vae-fp16-fix
Lastly, I would love to see these 3 models
Ri-Mix > https://civitai.com/models/996495/ri-mix-pony-illustrious?modelVersionId=1916444
BeyondReality > https://civitai.com/models/2455942/gs-or-beyond-reality-or-20
GoldenHour > https://civitai.com/models/2433139/golden-hour?modelVersionId=2735786
Thanks!
You can use any clip and vae that you want, and some good combinations can be made.
I've been using this VAE - https://civitai.com/models/2311905?modelVersionId=2601318
I'll convert them and will update you one they are available
@Santodan Thank you! My regular vae give very weird results before for some reason but now it doesn't, idk why :/
Btw I just realized but can you also update NovaOrange as well please? Its few months old version and don't have Clips
https://civitai.com/models/967405?modelVersionId=2784048
@jfmlol I've uploaded the other three, check my hugginface, I might have there.
but will check that new one
@Santodan Tbh I later realized that v13-14 and EX v1-2 are actually different. So v13 isn't that old as well.. Up to you.
Thanks for others!
@jfmlol both already available
@Santodan May I ask something? I am just trying to understand GGuF's better. From what I can see I can use other model clips as well without any issue. I can't see much visual difference either, why is that?
@jfmlol they are all based on SDXL, the difference will not much, what really make the difference is the model itself