Base Dockerfiles for ComfyUI

This repo contains a collection of dockerfiles for running ComfyUI with a particular version and flavor of CUDA.

It is meant to be used alongside my ComfyUI Environment Manager tool

Docker images built and pushed to Dockerhub and can be found here

Note: I am not a security expert and am open to Issues and PRs that can help to harden the Images included in this repo.

Updates

(06/20/2025):

• (v1.1.1) Add build-essential to image to fix sage-attention errors.
• (v1.1.0) Force comfy user to 1000:1000 to match default WSL distro user permissions. Added permissions checking on initial container startup, as well as a fix-permissions script which can be run via exec into the container (use comfydock dev exec for a convienient way to do so) and running the command fix-permissions.

(06/13/2025): Changed image tag format for new images to include comfydock image version number alongside comfyui version (will make it easier to release updates more frequently).

(06/07/2025): Removed CUDA 11.8 support for new images (Comfy versions v0.3.40+) due to incompatibilities with xformers and pytorch. Added full CUDA 12.8 support. Fixed updating ComfyUI via manager inside containers.

(04/03/2025): Increased security of docker images by switching from 'root' user inside containers to 'comfy' user. Modified permissions of /app directory to enable full read/write for all users. Available for images v0.3.29+.

Testing

To build the image locally, run the following command:

docker build \
  --build-arg COMFYUI_VERSION=master \
  --build-arg CUDA_VERSION=12.4 \
  --build-arg PYTHON_VERSION=3.12 \
  --build-arg PYTORCH_VERSION=stable \
  --tag comfydock-env:master-py3.12-cuda12.4 \
  --file Dockerfile \
  .

To run the image, run the following command:

docker run -it --rm --gpus all comfydock-env:master-py3.12-cuda12.4