ChiefNakor/comfyui-blackwell-docker

Containerizes ComfyUI with native NVFP4 quantization support, enabling 3x faster inference and 3.5x reduced VRAM consumption on Blackwell GPUs through hardware-accelerated 4-bit floating-point precision. Provides persistent volume mounting for models, outputs, and custom nodes while automating dependency resolution during Docker rebuilds—critical for custom node installation where `requirements.txt` files must be baked into the image. Integrates with ComfyUI Manager for node discovery and includes pre-configured Nunchaku wheel support plus optional SageAttention (v2/v3) optimization, with flexible `.env`-based configuration for GPU memory management, PyTorch/CUDA versions, and model paths.