ComfyUI-GGUF

Why we included this project

Most diffusion workloads hit a wall on mid-range and low-VRAM GPUs, and this ComfyUI extension is one way around that. Its loader nodes read model files in the GGUF quantized format that llama.cpp made popular, which lets you trade some image quality for a much smaller memory footprint. Transformer-based models like FLUX and Stable Diffusion 3.5 tolerate quantization better than the older UNet designs, and the pack also loads a quantized T5 text encoder, so the savings cover both the model weights and the conditioning text. Installation is a git clone into ComfyUI's custom_nodes folder with one extra pip dependency, and the GGUF loader replaces the stock diffusion model loader in existing workflows. The project still calls itself work in progress, so expect rough edges, but it has become a common choice for running large diffusion models on consumer hardware.