ai-dock ComfyUI

AI-Dock provides ComfyUI images for different hardware backends. Choose a tag based on your target platform: NVIDIA CUDA, AMD ROCm, or CPU.

Tag patterns:

• CUDA: :cuda-[x.x.x-base|runtime]-[ubuntu-version]
• ROCm: :rocm-[x.x.x-runtime]-[ubuntu-version]
• CPU: :cpu-[ubuntu-version]

The :latest tag points to :latest-cuda and represents a stable, tested version.

You can customize the ComfyUI container behavior using several environment variables. Note that these images do not bundle models; use a provisioning script to download them.

To enable hardware acceleration, you must configure the docker-compose.yaml file based on your GPU type.

NVIDIA GPUs

Uncomment the deploy section in docker-compose.yaml to allow the container to access all available NVIDIA GPUs:

deploy:
  resources:
    reservations:
      devices:
        - driver: nvidia
          count: all
          capabilities: [gpu]

AMD GPUs

For AMD hardware, ensure the device mapping for /dev/kfd is enabled in the devices section:

devices:
  - "/dev/dri:/dev/dri"
  - "/dev/kfd:/dev/kfd"

The ComfyUI API Wrapper is a service available on port 8188. It is intended to replace older serverless handlers.

• Direct API Access: /ai-dock/api/
• Swagger/OpenAPI Documentation: /ai-dock/api/docs

Note: All services are password protected by default.

The ComfyUI Docker image can be customized during the build process using environment variables in a .env file. These variables control the Python version, PyTorch version, and the base image used for the build.

Available build arguments:

• PYTHON_VERSION: The Python version to use (defaults to 3.10).
• PYTORCH_VERSION: The PyTorch version to use (defaults to 2.4.1).
• COMFYUI_BUILD_REF: Specific build reference for ComfyUI.
• IMAGE_BASE: The base image URL (defaults to ghcr.io/ai-dock/python:${PYTHON_VERSION}-v2-cuda-12.1.1-base-22.04).
• IMAGE_TAG: The tag for the resulting image.

# Example .env content
PYTHON_VERSION=3.10
PYTORCH_VERSION=2.4.1
IMAGE_TAG=custom-tag

The project uses a workspace volume to persist data. By default, the local ./workspace directory is mapped to the container's ${WORKSPACE} (defaulting to /workspace/).

To share storage with other non-ComfyUI containers, you can map a common storage path to the /workspace/storage/ subdirectory within the container:

volumes:
  - ./workspace:${WORKSPACE:-/workspace/}:rshared
  - /path/to/common_storage:${WORKSPACE:-/workspace/}storage/:rshared

The image includes pre-configured Micromamba environments. The comfyui environment is activated automatically upon shell login.

• comfyui: Contains ComfyUI and its dependencies.
• api: Contains the ComfyUI API wrapper and its dependencies.

ComfyUI runs as a service on port 8188 (unless COMFYUI_PORT_HOST is set). You can pass startup flags via the COMFYUI_ARGS environment variable.

Use supervisorctl to manage the service lifecycle:

# Examples of service management
supervisorctl start comfyui
supervisorctl stop comfyui
supervisorctl restart comfyui

The following environment variables control the host-side ports for various services. Use these to avoid conflicts with existing services on your machine: