How to Deploy ComfyUI Docker Image to Production Environment

Hello everyone, I am the daydreamer Nexmoe. Recently, I have open-sourced a ComfyUI elastic Serverless application packaged based on Docker, which features a complete front-end and back-end separation architecture and a user-friendly interface.

After completing the development of the ComfyUI workflow, deploying it to a production environment can be quite tricky. Therefore, I have open-sourced a set of examples for everyone to learn from.

Demo: https://hadoop.nexmoe.com/
Open-source address: https://github.com/nexmoe/serverless-comfyui

Project Features#

🐳 Complete Docker deployment solution
🎨 Modern front-end interface
🔌 Modular back-end architecture
🛠 Simple configuration and usage

Architecture Diagram#

Mermaid Loading...

Project Structure#

comfy-docker/
├── frontend/           # Next.js front-end project
│   ├── src/           # Source code
│   └── .env          # Environment configuration
├── backend/           # ComfyUI backend
│   ├── checkpoints/   # Model checkpoints
│   ├── controlnet/    # ControlNet models
│   ├── custom_nodes/  # Custom nodes
│   └── loras/        # LoRA models
└── bruno/            # API test files

The directory structure of frontend/ is as follows, models and custom nodes need to be downloaded and installed manually.

.
├── Dockerfile
├── checkpoints
│   └── dreamshaperXL_sfwV2TurboDPMSDE.safetensors
├── controlnet
│   ├── sai_xl_canny_256lora.safetensors
│   └── sai_xl_depth_256lora.safetensors
├── custom_nodes
│   ├── ComfyUI-Custom-Scripts
│   ├── ComfyUI-WD14-Tagger
│   ├── ComfyUI_Comfyroll_CustomNodes
│   ├── comfyui-art-venture
│   └── comfyui_controlnet_aux
├── docker-compose.yml
├── loras
│   └── StudioGhibli.Redmond-StdGBRRedmAF-StudioGhibli.safetensors
├── provisioning.sh  // Custom script
└── sanhua.json  // Workflow

Environment Requirements#

Docker & Docker Compose
NVIDIA GPU (current demo workflow requires more than 12G of VRAM)
Sufficient disk space (100G~200G) for storing models

Quick Start#

Local Testing of Backend#

Navigate to the backend Dockerfile directory

cd backend

Download model files

Please refer to: https://gongjiyun.com/docs/tutorials/comfyui.html#%E4%B8%8B%E8%BD%BD%E6%A8%A1%E5%9E%8B%E5%92%8C%E6%89%A9%E5%B1%95

Build the Docker image

docker build -t gongji/comfyui:0.1 .

Run the Docker container

docker run -it --rm --gpus all -p 3000:3000 -p 8188:8188 --name comfyui gongji/comfyui:0.1

After the container starts, you can access:

ComfyUI interface: http://localhost:8188
API interface: http://localhost:3000/docs

Local Testing of Frontend#

Navigate to the frontend directory

cd frontend

Configure environment variables

cp .env.example .env
# Edit the .env file to configure necessary environment variables

Install dependencies and start

pnpm install
pnpm dev

Deploying ComfyUI Docker to Serverless Elastic Platform#

Please refer to Gongji Technology's ComfyUI deployment documentation

API Documentation#

The project uses Bruno for API testing and documentation management, with related files located in the bruno/ directory.

Example of ComfyUI API Call#

Here is an example code for calling the ComfyUI API (refer to frontend/src/app/api/route.ts):

async function generateImage(imageUrl: string) {
    // 1. Prepare prompt data
    const promptData = { ...promptob };  // Import base prompt from JSON file
    promptData.prompt["30"].inputs.image = imageUrl;  // Modify input image

    // 2. Set request options
    const url = `${process.env.GONGJI_ENDPOINT}/prompt`;
    const options = {
        method: 'POST',
        headers: { 'content-type': 'application/json' },
        body: JSON.stringify(promptData)
    };

    // 3. Send request
    const response = await fetch(url, options);
    const data = await response.json();

    // 4. Error handling
    if (response.status !== 200) {
        throw new Error(response.statusText);
    }

    // 5. Process returned image data
    if (data.images && data.images.length > 0) {
        return data.images[0];  // Return base64 formatted image data
    } else {
        throw new Error('No valid image data returned');
    }
}

Main steps explanation:

Prepare Prompt:
- Import base prompt configuration from JSON file
- Modify parameters in the prompt as needed (e.g., input image)
Send Request:
- Use POST method
- Set Content-Type to application/json
- Request body is serialized prompt data
Process Response:
- Check response status code
- Parse returned JSON data
- Extract generated image (base64 format)
Error Handling:
- Log error messages
- Throw appropriate error messages

Environment Variable Configuration#

Before using the API, ensure to configure the following environment variables:

GONGJI_ENDPOINT=your-comfyui-api-endpoint  # ComfyUI API endpoint

S3 Configuration Instructions#

The project's image upload feature requires configuring S3 storage services. You can use AWS S3 or other object storage services compatible with S3 protocol (like MinIO).

Configure the following environment variables in the frontend/.env file:

S3_ENDPOINT=your-s3-endpoint
S3_ACCESS_KEY=your-access-key
S3_SECRET_KEY=your-secret-key
S3_BUCKET=your-bucket-name
S3_REGION=your-region

Note:

Ensure the created bucket has appropriate access permissions
If using MinIO, the endpoint should be a complete URL (e.g., http://localhost:9000)
When using AWS S3, the endpoint configuration can be omitted

Contribution Guidelines#

Feel free to submit Issues and Pull Requests!

License#

MIT License