ComfyUI-GGUF

Running FLUX-class models on modest hardware is the point of ComfyUI-GGUF, a set of custom nodes by city96 that brings the GGUF quantization format from the llama.cpp ecosystem to native ComfyUI models. It targets transformer and DiT architectures such as FLUX.1-dev, FLUX.1-schnell, and Stable Diffusion 3.5 large, which tolerate low-bit quantization far better than older UNet designs, and it also loads quantized T5-XXL text encoders to claw back further memory. Installation means cloning the repository into ComfyUI's custom_nodes folder and installing the gguf Python package, after which quantized checkpoints drop into the standard model folders and load through dedicated loader nodes; city96's prequantized GGUF conversions on Hugging Face are among the most widely used files for these models. Tooling for producing custom quantizations lives in the repository's tools folder. The project is Apache-2.0 licensed with about 3,900 stars, and it has become the default route for running FLUX and other large diffusion transformers on GPUs with limited VRAM.