Deploy ComfyUI with Docker for Local AI Image Generation

Run ComfyUI — the node-based AI image generation engine — in a Docker container on your own GPU. Generate unlimited images with Stable Diffusion, Flux, or Qwen-Image models with zero per-image cost, zero API dependencies, and zero prompts leaving your machine. This is the setup SMF Works uses for all brand visual production.

• A Dockerized ComfyUI server running on your GPU
• A web interface accessible at http://localhost:8188
• Support for SDXL, Flux.1, Qwen-Image, and other open checkpoints
• A persistent models directory that survives container restarts
• API access for programmatic image generation from agents and scripts

1. Create a persistent directory structure

mkdir -p ~/comfyui/{models,output,custom-nodes}
mkdir -p ~/comfyui/models/{checkpoints,loras,vae,clip,clip_vision,unet}

This keeps your model weights and outputs outside the container so they survive restarts.

2. Pull and run the ComfyUI Docker image

The official ComfyUI Docker image handles CUDA setup automatically:

docker run -d \
  --name comfyui \
  --gpus all \
  -p 8188:8188 \
  -v ~/comfyui/models:/app/models \
  -v ~/comfyui/output:/app/output \
  -v ~/comfyui/custom-nodes:/app/custom-nodes \
  --restart unless-stopped \
  yanwk/comfyui-boot:latest

Adjust the image tag to the current version. Check Docker Hub or the ComfyUI GitHub releases for the latest.

3. Download a model checkpoint

Download a model into your persistent models directory. For Flux.1 Schnell (fast, lower quality, Apache-2 license):

# Download into the unet directory
wget -O ~/comfyui/models/unet/flux1-schnell.safetensors \
  https://huggingface.co/black-forest-labs/FLUX.1-schnell/resolve/main/flux1-schnell.safetensors

For Stable Diffusion XL (smaller VRAM requirement):

wget -O ~/comfyui/models/checkpoints/sd_xl_base_1.0.safetensors \
  https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0/resolve/main/sd_xl_base_1.0.safetensors

4. Verify the server is running

Open http://localhost:8188 in your browser. You should see the ComfyUI node-based interface with a default workflow loaded.

5. Generate your first image

The default text-to-image workflow should load automatically. If not, go to Workflows > Templates and select a text-to-image template. Enter a prompt, click "Queue Prompt," and wait for the image to appear.

6. Connect an agent or script

ComfyUI exposes a REST API at http://localhost:8188/api. You can queue prompts programmatically:

curl -X POST http://localhost:8188/api/prompt \
  -H 'Content-Type: application/json' \
  -d '{"prompt": { ... workflow JSON ... }}'

For agent integration, the ComfyUI MCP server provides tool-based access. Hermes Agent can call ComfyUI through the MCP interface to generate images as part of a larger workflow.

• Browser interface loads at http://localhost:8188
• A test image generates successfully within 30-60 seconds (depending on GPU and model)
• docker logs comfyui shows no CUDA errors
• nvidia-smi inside the container shows the GPU is recognized

• GPU not detected: Verify NVIDIA Container Toolkit is installed (nvidia-ctk --version). Run docker run --rm --gpus all nvidia/cuda:12.0-base nvidia-smi to confirm Docker can see the GPU.
• OOM errors: Use a smaller model (SDXL instead of Flux) or reduce the image resolution. For Qwen-Image, ensure at least 14-16GB VRAM headroom.
• Model not found: Check that model files are in the correct subdirectory of ~/comfyui/models/. ComfyUI looks for checkpoints in models/checkpoints/, unets in models/unet/, etc.
• Slow generation: Confirm the GPU is actually being used — check nvidia-smi during generation. If CPU is being used instead, the container's CUDA setup is wrong.

ComfyUI's Docker image and node ecosystem change frequently. Custom nodes from ComfyUI Manager may have dependencies that break across versions. Pin your image tag and test node updates before deploying. The node-based interface has a learning curve — start with templates, do not build a workflow from scratch on day one.