Skip to content

ComfyUI-GGUF

Run quantised GGUF models inside ComfyUI

Visit official site

What ComfyUI-GGUF is

The project is described as adding GGUF quantisation support to native ComfyUI models, letting model files be stored in the GGUF format popularised by llama.cpp.

What you can do with it

  • Store model files in GGUF format with quantisation support
  • Run transformer and DiT models such as flux on low-end GPUs
  • Use custom nodes provided for model quantisation
  • Follow the included installation and usage guide

Application scenarios

  • Developers deploy flux on a low-end GPU to optimise resources
  • Researchers apply GGUF quantisation to models on edge devices
  • Educational institutions use the project as a deep learning optimisation case

Getting started

  • Confirm the ComfyUI version supports custom operations
  • Clone the ComfyUI-GGUF repository with git
  • Install the inference dependency with pip install --upgrade gguf
  • Place the .gguf model file in the ComfyUI/models/unet folder

Pros & cons

✓ What we like

  • Model files can be stored in the GGUF format
  • An installation and usage guide is included

! What to watch out for

  • LoRA and Controlnet support is not included because weights are quantised
  • Conventional UNET models using conv2d are stated as unsuited to quantisation

FAQ

Which models suit quantisation?

The material names transformer and DiT models such as flux, while conventional UNET conv2d models are described as unsuited.

Are LoRA and Controlnet supported?

The feature list states they are not included because the weights are already quantised.

Last reviewed: 2026-11-15

More AI image generation tools

View all →

How we review