ComfyUI Docker

ComfyUI Flux is a Docker-based setup for running ComfyUI with FLUX.1 models and HunyuanVideo as well as a few additional features.

Features

• Dockerized ComfyUI environment
• Automatic installation of ComfyUI and ComfyUI-Manager
• Low VRAM Mode: Download and use FP8 models for reduced VRAM usage
• Pre-configured with FLUX models and VAEs
• Ability to automatically install HunyuanVideo models and VAEs
• Easy model management and updates
• GPU support with CUDA 12.1 or 12.4

Prerequisites

• Docker and Docker Compose
• NVIDIA GPU with CUDA support (for GPU acceleration)
• (Optional) Huggingface account and token (for downloading FLUX.1[dev] official model)

Quick Start

1. (Optional) Create a .env file in the project root and add your Huggingface token:

   HF_TOKEN=your_huggingface_token
   LOW_VRAM=false  # Set to true to enable low VRAM mode
   COMFYUI_DOWNLOAD_VIDEO_MODELS=true # Download HunyuanVideo_repackaged related models (they are large)

2. Download the docker-compose.yml file:

wget https://raw.githubusercontent.com/ravensorb/comfyui-docker/main/compose.yml

Alternatively, you can create a compose.yml file and copy/paste the following contents:

services:
  comfyui:
    container_name: comfyui
    image: ravensorb/comfyui-docker:latest
    restart: unless-stopped
    ports:
      - "8188:8188"
    volumes:
      - "./data/app:/app"
      - "./data/data:/data"
      - "./data/config:/config"
    environment:
      - CLI_ARGS=
      - HF_TOKEN=${HF_TOKEN}
      - LOW_VRAM=${LOW_VRAM:-false}
      - COMFYUI_DOWNLOAD_VIDEO_MODELS=${COMFYUI_DOWNLOAD_VIDEO_MODELS:-true}
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities:
                - gpu

3. Run the container using Docker Compose:

   docker compose up -d

   Note: The first time you run the container, it will download all the included models before starting up. This process may take some time depending on your internet connection (could be 20 minutes or more).

4. Access ComfyUI in your browser at http://localhost:8188

Low VRAM Mode

By setting the LOW_VRAM environment variable to true, the container will download and use the FP8 models, which are optimized for lower VRAM usage. The FP8 versions have CLIP and VAE merged, so only the checkpoint files are needed.

Enable Low VRAM Mode:

LOW_VRAM=true

Model Files

Overview of the model files that will be automatically downloaded when using this container. Some model files require an HF_TOKEN for download.

When LOW_VRAM=false (default)

Type	Model File Name	Size	Notes
UNet	flux1-schnell.safetensors	23 GiB
UNet	flux1-dev.safetensors	23 GiB	requires HF_TOKEN for download
CLIP	clip_l.safetensors	235 MiB
CLIP	t5xxl_fp16.safetensors	9.2 GiB
CLIP	t5xxl_fp8_e4m3fn.safetensors	4.6 GiB
LoRA	flux_realism_lora.safetensors	22 MiB
VAE	ae.safetensors	320 MiB

When LOW_VRAM=true

Type	Model File Name	Size	Notes
Checkpoint	flux1-dev-fp8.safetensors	17 GiB
Checkpoint	flux1-schnell-fp8.safetensors	17 GiB

Workflows

Download the images below and drag them into ComfyUI to load the corresponding workflows.

Official versions

FLUX.1[schnell]	FLUX.1[dev]
Download	Download

FP8 versions (LOW_VRAM)

FLUX.1[schnell] FP8	FLUX.1[dev] FP8
Download	Download

Updating

The ComfyUI and ComfyUI-Manager are automatically updated when the container starts. To update the base image and other dependencies, pull the latest version of the Docker image using:

docker compose pull

Additional Notes

• Switching Between Modes: If you change the LOW_VRAM setting after the initial run, the container will automatically download the required models for the new setting upon restart.
• Model Downloading: The scripts are designed to skip downloading models that already exist, so you won't waste bandwidth re-downloading models you already have.
• Huggingface Token: The HF_TOKEN is only necessary for downloading the flux1-dev.safetensors model when LOW_VRAM=false.
• Customizing folders for ComfyUI: https://medium.com/xrlo-extended-reality-lowdown/sharing-models-and-custom-nodes-in-comfyui-0965ef7f1485
• ComfyUI Folder Structure: https://comfyui-wiki.com/en/interface/files