Innovate Futures @ Benji
patreon
Quick Tips - Flux 1 Dev using GGUF Quantization
🕑 Added 2024-09-02 17:36:15 +0000 UTCHello everyone!
So ComfyUI have a new custom node recently.
It allow ComfyUI (latest version) to support GGUF Quantization AI model.
Some Info. about GGUF Quantization :
ComfyUI-GGUF : https://github.com/city96/ComfyUI-GGUF
GGUF format popularized by llama.cpp.
And using a diffusion model like Flux , it helps to reduce the VRAM usage in the execution.
This is one of the image generation processes.
This only things you have to change on Flux workflow, Unet Loader(GGUF) and DualCLIPLoader(GGUF)
Here you can download the GGUF format models:
FLUX 1 Dev - GGUF
https://huggingface.co/city96/FLUX.1-dev-gguf
T5 Clip - GGUF
https://huggingface.co/city96/t5-v1_1-xxl-encoder-gguf
Currently I use Q8_0 quantization model, image quality same as FLUX 1 Dev FP8 so far.
For ComfyUI-GGUF Installation:
After custom node install in ComfyUI Manager.
for portable comfyui
.\python_embeded\python.exe -s -m pip install -r .\ComfyUI\custom_nodes\ComfyUI-GGUF\requirements.txt
For ComfyUI with system Python:
pip install --upgrade gguf
Simple TXT2IMG workflow attached below for testing.
Have fun! :)
