w-club/comfyui_docker

This is a personal, Docker-based experimental environment for ComfyUI.
It utilizes a high-performance base image with PyTorch 2.9.1 and CUDA 13.0, pre-integrated with essential and powerful custom node dependencies.

This build is specifically optimized to resolve complex dependency conflicts (“dependency hell”), providing an out-of-the-box experience.

---

📋 Prerequisites

Before running this container, ensure your host machine meets the following requirements:

• NVIDIA GPU with up-to-date drivers installed
• Docker and Docker Compose installed
• NVIDIA Container Toolkit installed and configured
  • This is strictly required for the container to access your GPU.
  • NVIDIA Container Toolkit – Install Guide

---

📦 Core Components

ComfyUI

The most powerful and modular Stable Diffusion GUI.

ComfyUI-Manager

Essential extension for managing custom nodes and models directly from the UI.

PyTorch Docker Image

Base environment running PyTorch 2.9.1 with CUDA 13.0 runtime.

---

🧩 Pre-installed Requirements for Custom Nodes

Important
This Docker image pre-installs the heavy Python dependencies (requirements) for the following nodes to ensure stability and avoid build errors (such as rsa or google-cloud conflicts).

You still need to search for and install these nodes themselves via ComfyUI-Manager inside the UI to complete the setup.

Pre-resolved Python requirements are included for:

• comfyui_controlnet_aux
  Comprehensive set of preprocessors for ControlNet.

• ComfyUI-KJNodes
  A suite of useful utility nodes for various workflows.

• comfyui-tensorops
  Tools for advanced tensor operations.

• was-node-suite-comfyui
  Extensive node suite for image processing and auxiliary tools.

• ComfyUI-Crystools
  Hardware monitoring (VRAM/RAM/GPU usage) and metadata readers.

• DZ-FaceDetailer
  Essential tool for face restoration and detailing.

• ComfyUI-Easy-Use
  Easy control for mutiple nodes.

• comfyui_LLM_party
  Integration for Large Language Models (OpenAI, Claude, etc.).

Note
The aisuite dependency has been modified to use the core package instead of the full aisuite[all] variant to avoid conflicts with Google Cloud SDKs.

---

📂 Suggested Host Folder Structure

You can organize your host machine like this:

.
├── docker-compose.yml
├── comfyui/          # ComfyUI core from container
└── ai/
    ├── models/       # Models: checkpoints, LoRA, VAE, etc.
    ├── output/       # Generated images
    ├── input/        # Source images / assets
    └── user/         # Workflows, custom configs, user data

These paths are mapped into the container via Docker volumes (see below).

---

🚀 Quick Start

Use Docker Compose to spin up the service.

docker-compose.yaml

services:
  comfyui:
    # Always use the latest stable build
    image: wclu6/comfyui_docker:latest
    container_name: comfyui

    ports:
      - "8188:8188"

    volumes:
      # Map the ComfyUI core directory
      - ./comfyui:/comfyui

      # Models storage (Checkpoints, LoRAs, VAEs, etc.)
      - ./ai/models:/comfyui/models

      # Output directory for generated images
      - ./ai/output:/comfyui/output

      # Input directory for source images
      - ./ai/input:/comfyui/input

      # User directory for custom settings and workflows
      - ./ai/user:/comfyui/user

    deploy:
      resources:
        reservations:
          devices:
            - capabilities:
                - gpu

    environment:
      - NVIDIA_VISIBLE_DEVICES=all
      # Set to 'true' for multi-user mode, or 'false' for single user. Configurable.
      - MULTI_USER=true

    restart: unless-stopped

🚀 Launch Command

Run the following command in your terminal:

docker compose up -d

Once started, access the interface at:

http://localhost:8188

🛠️ Build Information

This image has been customized to address specific dependency issues found in comfyui_LLM_party and several heavy third-party libraries.

Dependency Fix

• Switched aisuite[all] → aisuite core package
• Avoids version conflicts involving:
  • rsa
  • google-auth
  • google-cloud-*
• Ensures a successful build and more stable runtime.

---

🔍 Notes & Requirements

• Required: NVIDIA drivers and nvidia-container-toolkit must be correctly installed on the host, otherwise GPU access will not work.
• Required: The ./comfyui:/comfyui bind mount must be present for this setup, as the container expects ComfyUI to live at /comfyui inside the container.
• You can still install additional custom nodes via ComfyUI-Manager inside the UI.
• Ensure you have enough VRAM for your chosen models and workflows.