Summary

This Docker image provides a pre-configured, isolated environment for running ComfyUI on NVIDIA GPUs. It targets users needing robust host OS separation, simplified setup, and reliable GPU acceleration. Benefits include a permission-aware ComfyUI deployment optimized for NVIDIA hardware, with ComfyUI-Manager for easy updates.

How It Works

Leveraging official NVIDIA CUDA base images, it runs ComfyUI as a non-root user with configurable UID/GID mapping for host file permission compatibility. Distinct run (code, venv) and basedir (models, user data) volumes ensure modularity. ComfyUI-Manager is pre-installed, and the environment supports multiple Ubuntu/CUDA versions for driver compatibility. It also integrates the uv package manager for faster Python dependency installation.

Quick Start & Requirements

• Install/Run: Use docker run or podman run with NVIDIA Container Toolkit, mounting run and basedir directories. Example commands are provided.
• Prerequisites: NVIDIA drivers compatible with chosen CUDA version, NVIDIA Container Toolkit.
• Resource Footprint: ~15GB for container/venv, excluding user data.
• Docs: NVIDIA Docker/Podman setup: https://www.gkr.one/blg-20240523-u24-nvidia-docker-podman.

Highlighted Details

• Version Flexibility: Numerous Ubuntu + CUDA tags (e.g., ubuntu24_cuda12.6.3, ubuntu24_cuda12.8) ensure broad NVIDIA driver compatibility.
• User Permissions: Runs as non-root comfy user, mapping host UID/GID via WANTED_UID/WANTED_GID for correct file ownership.
• Extensibility: Supports custom scripts (user_script.bash, /userscripts_dir) for advanced installations.
• ComfyUI-Manager: Pre-installed for easy updates and management, with configurable security levels.
• Directory Separation: Distinct run (code, venv) and basedir (models, user files) volumes enhance organization.
• uv Support: Enables uv for faster Python package management (USE_UV=true).

Maintenance & Community

The primary maintainer is mmartial. No specific community channels or detailed contributor information are listed.

Licensing & Compatibility

• License: Governed by the NVIDIA Deep Learning Container License.
• Compatibility: NVIDIA GPUs, Docker, Podman, WSL2 on Windows, various Ubuntu/CUDA versions.

Limitations & Caveats

Switching OS+CUDA versions may cause custom node import failures. Using BASE_DIRECTORY with outdated ComfyUI can lead to errors. Intended for self-hosted models, not API nodes. podman-compose may not function on WSL2.