ComfyUI Docker Quickstart

This guide details containerizing and running ComfyUI with Docker on Linux and Windows (including WSL2). It covers mounting local directories, best practices for I/O performance, and an Ollama integration section.

Pre-requisites
Selecting the Appropriate Docker Image
Linux: Running ComfyUI in Docker
Windows: Running ComfyUI in Docker
Optional: Configure WSL2 Memory (Windows Only)
Optional: Mounting External Drives (Models/Output)
Ollama Integration with ComfyUI

Pre-requisites

Docker Desktop installed.
NVIDIA drivers + NVIDIA Container Toolkit on Linux. (For Windows using WSL2, Docker Desktop manages GPU support—no separate toolkit required.)
Full CUDA toolkit installation is unnecessary.

Selecting the Appropriate Docker Image

Default: NVIDIA GPU image yanwk/comfyui-boot:cu124-slim. For other GPUs, consult the available tags.

Linux: Running ComfyUI in Docker

mkdir -p comfy
docker run -it \
  --name comfyui-cu124 \
  --gpus all \
  -p 8188:8188 \
  -v "$(pwd)"/comfy:/root \
  -e CLI_ARGS="" \
  yanwk/comfyui-boot:cu124-slim

Access: http://localhost:8188/

Windows: Running ComfyUI in Docker

New-Item -ItemType Directory -Force -Path comfy
docker run -it `
  --name comfyui-cu124 `
  --gpus all `
  -p 8188:8188 `
  -v "${PWD}\comfy:/root" `
  -e CLI_ARGS="" `
  yanwk/comfyui-boot:cu124-slim

Access: http://localhost:8188/

Optional: Configure WSL2 Memory (Windows Only)

Increase available RAM for WSL2:

Edit %USERPROFILE%\.wslconfig:
```
[wsl2]
memory=56GB
```
Shutdown WSL2:
```
wsl --shutdown
```
Restart Docker Desktop.

Optional: Mounting External Drives (Models/Output)

Note: Bind mounts from Windows/NTFS or external drives incur slow I/O due to filesystem translation. For optimal performance, use Docker volumes or store files in the native Linux filesystem.

Windows

docker run -it `
  --name comfyui-cu124 `
  --gpus all `
  -p 8188:8188 `
  -v "${PWD}\comfy:/root" `
  -v "D:/AI/models:/root/ComfyUI/models" `
  -v "D:/AI/output:/root/ComfyUI/output" `
  -e CLI_ARGS="" `
  yanwk/comfyui-boot:cu124-slim

Linux

docker run -it \
  --name comfyui-cu124 \
  --gpus all \
  -p 8188:8188 \
  -v "$(pwd)"/comfy:/root \
  -v "/mnt/AI/models:/root/ComfyUI/models" \
  -v "/mnt/AI/output:/root/ComfyUI/output" \
  -e CLI_ARGS="" \
  yanwk/comfyui-boot:cu124-slim

Ollama Integration with ComfyUI

For advanced model execution, integrate with ComfyUI Ollama. Steps:

Deploy Ollama:
Run Ollama in Docker using a dedicated volume (e.g., ollama_data).
Avoid bind mounts on Windows as they yield poor I/O performance.

Migrate Existing Data (if needed):
Use a tar archive to move data to the volume:

docker run --rm `
  -v ollama_data:/volume `
  -v "E:/backup:/backup" `
  alpine sh -c "cd /volume && tar czf /backup/ollama_data_backup.tar.gz ."

Restore by extracting into the volume.

Launch the ComfyUI Ollama Container:
Mount the ollama_data volume:

docker run -it `
  --name comfyui-ollama `
  --gpus all `
  -p 8189:8189 `
  -v "${PWD}\comfy:/root" `
  -v ollama_data:/root/.ollama `
  -e CLI_ARGS="" `
  your-ollama-image

Access & Configure:
Navigate to http://localhost:8189/ and follow the instructions on the ComfyUI Ollama GitHub repo.

Resources:

ph33nx/ComfyUI-Docker-Quickstart.md