ComfyUI-GGUF is a custom node pack that adds GGUF quantization support to ComfyUI for running quantized diffusion models on resource-constrained hardware.
The tool addresses the challenge of running large diffusion models on low-end GPUs by enabling GGUF quantization, a format popularized by llama.cpp. While traditional UNET models with convolutional layers resist quantization, transformer-based and DiT models like Flux tolerate lower bit-depth quantization with minimal quality loss. The custom nodes allow loading quantized UNET models and quantized T5 text encoders, reducing VRAM requirements substantially. Users place GGUF model files in the standard ComfyUI models directory and swap the stock diffusion model loader for the GGUF-specific loader node.
Adoption suits developers running ComfyUI on GPUs with limited VRAM who want to use modern transformer-based diffusion models. The project provides pre-quantized versions of Flux and Stable Diffusion 3.5 models, along with quantized T5 encoders, eliminating the need to quantize models yourself. The tool includes experimental LoRA loading support. Users should ensure their ComfyUI installation is recent enough to support custom operations when loading UNET models, and on MacOS Sequoia, torch 2.4.1 is required to avoid buffer size errors.
The project maintains active engagement with its user base, with nearly all open issues originating from adopters reporting real-world problems rather than from the core team. Maintainers typically respond to new issues and pull requests within a day, indicating responsive stewardship of the codebase.
