Settings:
V1:Euler, simple, cfg 1-1.5 (2 can become blurry), steps 6-10.
V2:Euler, simple, cfg 1-1.5, steps 4-10.
Interesting negative I sometimes like to try: perfect symmetric face,
Uses the new 4 and 8 step loras from https://huggingface.co/alibaba-pai/Z-Image-Fun-Lora-Distill/tree/main combined, plus my big lora.
Description
FAQ
Comments (12)
Can you please, please, please tell me how you created a Q8 from a ZIB model? All the methods I've used so far (City96, llama.cpp, ModelQuantizer node) fail with ZIB and Flux.2 Klein. The internet, Gemini, ChatGPT, etc. don't know the answer either and all give the same answers, none of which work.
the city96 tools work you might just have to use a newer branch like this one https://github.com/city96/ComfyUI-GGUF/tree/auto_convert
ggufy can convert lumina2 based models as well, including z-image turbo or base (disclaimer: I made this tool) https://github.com/qskousen/ggufy
@ferretduck Yeaaaah, thank you for your work! I will try it! Also Flux?
Edit: Linux and Windows only, I'll have to see if I can get it to run on MacOS somehow...
Edit 2: Unfortunately not. Tried to compile ggml-alloc.c (classic ggml allocator file), but it is missing from the vendor/ggml/src/ folder of the repository → FileNotFound → CacheCheckFailed.
@ferretduck nice I will have a look sometime
@heinerheinervonvielen417 sorry about that! I don't have a mac at the moment so I am unable to debug this currently.
@ferretduck No problem, maybe the default ggml-alloc.c File will work…?!
@heinerheinervonvielen417 I'll try to debug this in the next few days as I have time.
@ferretduck I'm happy to help you if I can.
@heinerheinervonvielen417 I've DM'd you so we can continue this somewhere other than this model's comment section, haha.
just in case anyone read this and wondered -- ggufy now works on MacOS, both arm64 (apple silicon, M-series) and x86_64 (intel)! thanks @heinerheinervonvielen417 for the help testing those!
@ferretduck This tool, ladies and gentlemen, is the fastest and easiest tool you will find! It was my pleasure to help!