ComfyUI Dockerfile overhaul, GGUF, XFLUX, and krita-ai-diffusion support

[lainedfles]

Hi, as requested, here is the discussion accompanying #739

Dockerfile overhaul

Enhance the Dockerfile for the ComfyUI service by introducing several new features and configurations. The ARGs provide flexibility in configuring the Docker container, allowing users to tailor the ComfyUI installation to their specific needs. Here's a detailed breakdown of the changes:

1. Base Image Update:

   • Updated the base image from pytorch/pytorch:2.3.0-cuda12.1-cudnn8-runtime to pytorch/pytorch:2.5.1-cuda12.4-cudnn9-runtime. This ensures compatibility with newer versions of PyTorch and CUDA, providing better performance and support for recent hardware.

2. Environment Variables:

   • Introduced several new environment variables (USE_UID, USE_GID, USE_USER, USE_GROUP) to allow customization of the user and group IDs within the container.
   • Added flags (USE_EDGE, USE_GGUF, USE_XFLUX, USE_CNAUX, USE_KRITA, USE_IPAPLUS, USE_INPAINT, USE_TOOLING) to enable or disable optional features and integrations.

3. User/Group Management:

   • Added logic to create a user and group with specified IDs (USE_UID and USE_GID). This is useful for running the container in environments where specific user permissions are required.
   • Set default values for these variables, ensuring that the Dockerfile remains functional without explicit configuration.

4. Optional Feature Integration:

   • Included conditional logic to clone and install additional repositories based on the flags set (USE_GGUF, USE_XFLUX, USE_CNAUX, USE_KRITA, USE_IPAPLUS, USE_INPAINT, USE_TOOLING). This allows users to customize their ComfyUI installation by enabling only the features they need.
   • Ensured that dependencies for these optional features are installed correctly, including handling specific cases like separating ONNX Runtime installations to restore CUDA support.

5. Python Version Update:

   • Changed the Python command from python to python3 in the CMD instruction to explicitly specify the use of Python 3, which is a best practice for clarity and compatibility.

6. File Permissions and Ownership:

   • Used --chown=${USE_UID}:${USE_GID} when copying files into the container to ensure that the correct user owns these files, preventing permission issues during execution.

7. General Improvements:

   • Improved readability and maintainability of the Dockerfile by organizing the steps logically and adding comments where necessary.
   • Ensured that all commands are idempotent, meaning they can be run multiple times without causing unintended side effects.

• USE_UID: Specifies the user ID for the non-root user within the container. Default is set to value 0.
• USE_GID: Specifies the group ID for the non-root user within the container. Default is set to value like 0.
• USE_USER: Specifies the username for the non-root user within the container. Default is set to root.
• USE_GROUP: Specifies the group name for the non-root user within the container. Default is set to root.
• USE_EDGE: If set to true, clones and installs the latest development version of ComfyUI from the main branch.
• USE_GGUF: If set to true, clones and installs the ComfyUI-GGUF extension for GPU acceleration.
• USE_XFLUX: If set to true, clones and installs the x-flux-comfyui extension for additional functionalities.
• USE_CNAUX: If set to true, clones and installs the comfyui_controlnet_aux extension for control net auxiliary features.
• USE_KRITA: If set to true, clones and installs multiple extensions (comfyui_controlnet_aux, ComfyUI_IPAdapter_plus, comfyui-inpaint-nodes, comfyui-tooling-nodes) that are useful when integrating with Krita.
• USE_IPAPLUS: If set to true, clones and installs the ComfyUI_IPAdapter_plus extension for IP Adapter functionalities.
• USE_INPAINT: If set to true, clones and installs the comfyui-inpaint-nodes extension for inpainting capabilities.
• USE_TOOLING: If set to true, clones and installs the comfyui-tooling-nodes extension for additional tooling features.

New entrypoint.sh features

Introduce several enhancements to the entrypoint.sh script for the Comfy service. These changes enhance the script's flexibility, making it easier to integrate additional features and manage dependencies dynamically based on environment configurations:

1. Custom Node Update: Added logic to update custom nodes located in /data/config/comfy/custom_nodes/. If the UPDATE_CUSTOM_NODES environment variable is set to true, it will pull the latest changes from each node's Git repository.

2. Environment Variable Usage: Introduced a new environment variable CACHE which should be used instead of hard-coding /root/.cache. This makes the script more flexible and configurable.

3. Krita Integration: Added support for Krita integration with ComfyUI, including downloading models if KRITA_DOWNLOAD_MODELS is set to true, managing upscale models, and setting up symbolic links for model directories.

4. GGUF, XFLUX, CNAUX, IPAPLUS, INPAINT, TOOLING Support: Included logic to handle additional custom nodes (ComfyUI-GGUF, x-flux-comfyui, comfyui_controlnet_aux, ComfyUI_IPAdapter_plus, comfyui-inpaint-nodes, comfyui-tooling-nodes) based on their respective environment variables (USE_GGUF, USE_XFLUX, USE_CNAUX, USE_IPAPLUS, USE_INPAINT, USE_TOOLING). These nodes are copied to the custom nodes directory if they don't already exist.

.env support

Add .env file support to docker-compose file for environment variable management across all defined services in the docker-compose.yml. By adding env_file: .env, each service will now load environment variables from a local .env file, enhancing configuration flexibility and security. This change ensures that sensitive information or common settings can be centralized and easily managed without hard-coding them into the Docker Compose file.

Examples

Podman

Build and run container with one line enabling bleeding edge ComfyUi, enable GGUF, XFLUX, and KRITA support (download recommended models):

podman-compose --podman-build-args='--build-arg="USE_EDGE=true" --build-arg="USE_GGUF=true" --build-arg="USE_XFLUX=true" --build-arg="USE_KRITA=true"' --podman-run-args='-e KRITA_DOWNLOAD_MODELS=true --gpus all' --profile comfy up

Build a container suitable for use with a systemd service to run as non-root:

podman-compose --verbose --podman-build-args='--build-arg="USE_UID=991" --build-arg="USE_GID=991" --build-arg="USE_USER=app" --build-arg="USE_GROUP=app" --build-arg="USE_EDGE=true" --build-arg="USE_GGUF=true" --build-arg="USE_XFLUX=true" --build-arg="USE_KRITA=true"' --profile comfy build

Docker

Build to enable Krita support:

docker compose --profile comfy-cpu build --build-arg USE_KRITA=true

Use .env file to enable node updates and download recomended Krita models:

cat <<EOF> .env
UPDATE_CUSTOM_NODES=true
KRITA_DOWNLOAD_MODELS=true
EOF

Run container:

docker compose --profile comfy-cpu up

Disclaimer: no code is AI generated (just old-fashioned human AGI) but various local LLMs were used to author commit messages.