YanWenKun/ComfyUI-Docker

What it solves

It simplifies the installation and deployment of ComfyUI, a node-based GUI for AI-generated content (AIGC), by providing pre-configured Docker images. This removes the need for users to manually manage complex dependencies, CUDA versions, and environment setups across different hardware architectures.

How it works

The project provides a variety of Docker images tailored to specific hardware and needs:

• NVIDIA GPU images: Available in different CUDA versions (e.g., 13.0, 12.6) to match specific GPU architectures from Blackwell to Maxwell.
• AMD GPU images: Support for ROCm 6 and 7.
• Intel GPU images: Support for XPU.
• CPU-only images: A lightweight version for systems without a dedicated GPU.

Images are offered in two main flavors: "Slim" (containing ComfyUI and ComfyUI-Manager) and "MEGAPAK" (all-in-one bundles including development kits and numerous custom nodes).

Who it’s for

• Users who want to run ComfyUI without dealing with manual environment configuration.
• Developers and artists using NVIDIA, AMD, or Intel GPUs who need a stable, containerized environment.
• Beginners who prefer a "slim" setup or power users who want a pre-loaded "megapak" of custom nodes.

Highlights

• Broad Hardware Support: Dedicated images for NVIDIA, AMD (ROCm), Intel (XPU), and CPU.
• Architecture-Specific CUDA Tags: Clear mapping of CUDA versions to GPU architectures (e.g., RTX 5090, 4090, 3090).
• Flexible Image Options: Choice between lightweight slim images and comprehensive megapak bundles.
• Persistent Storage: Detailed volume mapping for models, custom nodes, and user profiles to ensure data persists outside the container.