Skip to content
/ comfyui-docker Public
forked from ashleykleynhans/comfyui-docker

Docker image for ComfyUI: The most powerful and modular stable diffusion GUI, api and backend with a graph/nodes interface.

License

GPL-3.0 license
0 stars 25 forks Branches Tags Activity
Star
Notifications You must be signed in to change notification settings

KilianKae/comfyui-docker

BranchesTags
 
 

Repository files navigation

ComfyUI Production Docker

A production-grade Docker setup for ComfyUI with automated model downloading, 50+ pre-installed custom nodes, and optimized for RunPod cloud deployment.

Features

  • Multi-Variant Builds: Support for CUDA 12.4/12.8 with Python 3.11/3.12
  • 50+ Custom Nodes: Pre-installed custom nodes including Login, KJNodes, RMBG, SAM2, JoyCaption, and more
  • Automated Model Downloading: Runtime model provisioning from Hugging Face, CivitAI, and direct URLs
  • Production Ready: NGINX reverse proxy, App Manager, persistent storage sync
  • RunPod Optimized: Intelligent syncing for RunPod network volumes
  • Flexible Configuration: Environment variable-based model provisioning

Quick Start

Building the Image

# Build default variant (CUDA 12.8, Python 3.12 - RTX 5090)
docker buildx bake

# Build specific variant (CUDA 12.4, Python 3.11 - RTX 4090, A6000)
docker buildx bake cu124-py311

# Build all variants
docker buildx bake all

Running the Container

Basic Usage

docker run -it --gpus all \
  -p 3000:3000 \
  -p 8000:8000 \
  -v $(pwd)/workspace:/workspace \
  kiliankaslin/comfyui-production:cu128-py312-v1.0.0

With Model Provisioning

docker run -it --gpus all \
  -p 3000:3000 \
  -p 8000:8000 \
  -v $(pwd)/workspace:/workspace \
  -e HF_MODEL_DIFFUSION_MODELS1=Comfy-Org/flux1-dev \
  -e HF_MODEL_DIFFUSION_MODELS_FILENAME1=flux1-dev.safetensors \
  -e HF_MODEL_TEXT_ENCODERS1=comfyanonymous/flux_text_encoders \
  -e HF_MODEL_TEXT_ENCODERS_FILENAME1=t5xxl_fp16.safetensors \
  -e CIVITAI_TOKEN=your_civitai_api_token \
  -e CIVITAI_MODEL_LORA_URL1=https://civitai.com/api/download/models/123456 \
  kiliankaslin/comfyui-production:cu128-py312-v1.0.0

Environment Variables

Service Configuration

Variable Description Default
DISABLE_AUTOLAUNCH Skip auto-starting ComfyUI Not set (auto-launch enabled)
DISABLE_SYNC Skip syncing to persistent storage Not set (sync enabled)
EXTRA_ARGS Additional ComfyUI arguments Not set
PUBLIC_KEY SSH public key for remote access Not set

Model Provisioning

Workflow Downloads

WORKFLOW1=https://example.com/workflow.json
WORKFLOW2=https://example.com/my-workflow.zip
# ... up to WORKFLOW50

Supports: .json, .zip, .tar.gz, .tar.xz, .tar.bz2, .7z

Hugging Face Models (Categorized)

Download models to specific ComfyUI model directories:

# VAE models
HF_MODEL_VAE1=stabilityai/sdxl-vae
HF_MODEL_VAE_FILENAME1=sdxl_vae.safetensors

# LoRA models
HF_MODEL_LORA1=your-repo/your-lora
HF_MODEL_LORA_FILENAME1=lora.safetensors

# Diffusion models
HF_MODEL_DIFFUSION_MODELS1=Comfy-Org/flux1-dev
HF_MODEL_DIFFUSION_MODELS_FILENAME1=flux1-dev.safetensors

# Text encoders
HF_MODEL_TEXT_ENCODERS1=comfyanonymous/flux_text_encoders
HF_MODEL_TEXT_ENCODERS_FILENAME1=t5xxl_fp16.safetensors

# Upscale models
HF_MODEL_UPSCALER1=your-repo/upscaler
HF_MODEL_UPSCALER_PTH1=model.pth

# CLIP Vision
HF_MODEL_CLIP_VISION1=your-repo/clip
HF_MODEL_CLIP_VISION_FILENAME1=clip.safetensors

# Model Patches
HF_MODEL_PATCHES1=your-repo/patch
HF_MODEL_PATCHES_FILENAME1=patch.safetensors

# Audio Encoders
HF_MODEL_AUDIO_ENCODERS1=your-repo/audio
HF_MODEL_AUDIO_ENCODERS_FILENAME1=audio.safetensors

# Checkpoints
HF_MODEL_CHECKPOINTS1=your-repo/checkpoint
HF_MODEL_CHECKPOINTS_FILENAME1=checkpoint.safetensors

# VLM (Vision-Language Models)
HF_MODEL_VL1=your-repo/vlm
HF_MODEL_VL_FILENAME1=vlm.safetensors

# SAM models
HF_MODEL_SAMS1=your-repo/sam
HF_MODEL_SAMS_FILENAME1=sam.safetensors

# Latent Upscale
HF_MODEL_LATENT_UPSCALE1=your-repo/latent
HF_MODEL_LATENT_UPSCALE_FILENAME1=latent.safetensors

Each category supports up to 20 indexed downloads (e.g., HF_MODEL_VAE1 through HF_MODEL_VAE20)

Generic Hugging Face Downloads

For custom paths:

# Download single file
HF_MODEL1=repo/model-name
HF_MODEL_FILENAME1=specific_file.safetensors
HF_MODEL_DIR1=models/custom_path

# Download full model
HF_FULL_MODEL1=repo/full-model
HF_MODEL_DIR1=models/another_path

# ... up to HF_MODEL20 / HF_FULL_MODEL20

CivitAI Downloads

CIVITAI_TOKEN=your_civitai_api_token
CIVITAI_MODEL_LORA_URL1=https://civitai.com/api/download/models/123456
CIVITAI_MODEL_LORA_URL2=https://civitai.com/api/download/models/789012
# ... up to CIVITAI_MODEL_LORA_URL50

Note: CIVITAI_TOKEN is required for CivitAI downloads.

Directory Structure

/workspace/ComfyUI/
├── models/
│   ├── vae/                      # VAE models
│   ├── upscale_models/           # Upscaler models
│   ├── loras/                    # LoRA fine-tuning models
│   ├── text_encoders/            # CLIP/Text encoders
│   ├── clip_vision/              # CLIP vision models
│   ├── model_patches/            # Model patches
│   ├── audio_encoders/           # Audio encoders
│   ├── diffusion_models/         # Main diffusion models
│   ├── checkpoints/              # Model checkpoints
│   ├── VLM/                      # Vision-Language Models
│   ├── sams/                     # SAM models
│   └── latent_upscale_models/    # Latent upscaling models
├── user/default/workflows/       # Workflow files
└── custom_nodes/                 # Custom nodes (50+ pre-installed)

Pre-installed Custom Nodes

This image includes 50+ custom nodes:

  • rgthree-comfy: Enhanced UI and utilities
  • ComfyUI-Login: Authentication and user management
  • ComfyUI-KJNodes: Extended node collection
  • ComfyUI-RMBG: Background removal
  • ComfyUI-segment-anything-2: SAM2 integration
  • ComfyUI-JoyCaption: Image captioning
  • ComfyUI_UltimateSDUpscale: Advanced upscaling
  • ComfyUI-VideoHelperSuite: Video processing
  • ComfyUI-Impact-Pack: Quality improvements
  • ComfyUI-GGUF: GGUF model support
  • comfyui_controlnet_aux: ControlNet preprocessing
  • ComfyUI-Lora-Manager: LoRA management
  • And 40+ more for various AI tasks

Service Ports

Service Port Description
ComfyUI 3000 Main ComfyUI interface (proxied via NGINX)
App Manager 8000 Application management dashboard

Internal ports (not exposed by default):

  • ComfyUI internal: 3001
  • Jupyter Lab: 8888
  • Code Server: 7777

Building Custom Variants

Modify docker-bake.hcl to customize builds:

variable "RELEASE" {
    default = "v1.0.0"  # Your version
}

variable "REGISTRY_USER" {
    default = "kiliankaslin"  # Your Docker Hub username
}

Advanced Configuration

Custom ComfyUI Arguments

docker run ... \
  -e EXTRA_ARGS="--lowvram --disable-xformers" \
  kiliankaslin/comfyui-production:tag

SSH Access

docker run ... \
  -p 22:22 \
  -e PUBLIC_KEY="ssh-rsa AAAA..." \
  kiliankaslin/comfyui-production:tag

Disable Persistent Storage Sync

docker run ... \
  -e DISABLE_SYNC=true \
  kiliankaslin/comfyui-production:tag

Example: Full FLUX Model Setup

docker run -it --gpus all \
  -p 3000:3000 \
  -p 8000:8000 \
  -v $(pwd)/workspace:/workspace \
  -e HF_MODEL_DIFFUSION_MODELS1=Comfy-Org/flux1-dev \
  -e HF_MODEL_DIFFUSION_MODELS_FILENAME1=flux1-dev.safetensors \
  -e HF_MODEL_TEXT_ENCODERS1=comfyanonymous/flux_text_encoders \
  -e HF_MODEL_TEXT_ENCODERS_FILENAME1=t5xxl_fp16.safetensors \
  -e HF_MODEL_TEXT_ENCODERS2=comfyanonymous/flux_text_encoders \
  -e HF_MODEL_TEXT_ENCODERS_FILENAME2=clip_l.safetensors \
  -e HF_MODEL_VAE1=black-forest-labs/FLUX.1-schnell \
  -e HF_MODEL_VAE_FILENAME1=ae.safetensors \
  -e WORKFLOW1=https://your-server.com/flux-workflow.json \
  kiliankaslin/comfyui-production:cu128-py312-v1.0.0

Troubleshooting

Models Not Downloading

  1. Check container logs: docker logs <container-id>
  2. Verify ComfyUI is online (wait 2-3 minutes after container start)
  3. Check environment variables are correctly set
  4. For CivitAI: Ensure CIVITAI_TOKEN is set

GPU Not Detected

  1. Verify NVIDIA drivers: nvidia-smi
  2. Check Docker GPU runtime: docker run --gpus all nvidia/cuda:12.4.1-base-ubuntu22.04 nvidia-smi
  3. Ensure --gpus all flag is used

Out of Memory

Use VRAM optimizations:

-e EXTRA_ARGS="--lowvram"
# or for very limited VRAM:
-e EXTRA_ARGS="--novram"

Architecture

This setup combines:

  • comfyui-docker: Production-grade base with NGINX, App Manager, persistent sync
  • run-comfyui-image: Runtime model provisioning capabilities
  • 50+ Custom Nodes: Extended functionality out of the box

License

This Docker setup integrates various open-source projects. Refer to individual component licenses:

  • ComfyUI: GPL-3.0
  • Custom nodes: Various (see respective repositories)

Credits

Built on top of:

Support

For issues specific to this Docker setup:

  • Check the troubleshooting section above
  • Review container logs
  • Ensure all prerequisites are met (GPU drivers, Docker, etc.)

For ComfyUI-specific issues:

About

Docker image for ComfyUI: The most powerful and modular stable diffusion GUI, api and backend with a graph/nodes interface.

Resources

Readme

License

GPL-3.0 license
Activity

Stars

0 stars

Watchers

0 watching

Forks

0 forks
Report repository

Releases

No releases published

Packages

No packages published

Languages

  • Shell 89.8%
  • HCL 7.6%
  • Dockerfile 2.6%