Custom Server¶

Run ComfyUI Studio on your own machine with Docker.

Prerequisites¶

NVIDIA GPU with CUDA support
Docker with NVIDIA Container Toolkit installed
At least 20 GB disk space for the Docker image
At least 50 GB additional space for AI models

Quick Start¶

Pull the Image¶

docker pull ghcr.io/diego-devita/comfyui-studio:latest

Run¶

docker run -d --gpus all \
  -p 8000:8000 \
  -p 8188:8188 \
  -v /path/to/data:/workspace \
  -e API_KEY=your-strong-password \
  -e CIVITAI_API_KEY=your-civitai-token \
  -e HF_TOKEN=your-hf-token \
  --name comfyui-studio \
  ghcr.io/diego-devita/comfyui-studio:latest

Replace /path/to/data with a directory on your host where you want persistent data (models, outputs, configuration).

Access¶

Open http://localhost:8000 and log in with your API_KEY.

Build Your Own Image¶

If the pre-built image doesn't match your GPU, use the build configurator:

git clone https://github.com/diego-devita/comfyui-studio.git
cd comfyui-studio
docker/configure.sh

The configurator asks about your GPU and outputs the right docker build command. See Build Configurator for details.

Environment Variables¶

Same as RunPod deployment. See Environment Variables for the full reference.

Key differences from RunPod:

Variable	RunPod	Custom Server
`STUDIO_DIR`	`/workspace/studio` (network volume)	Wherever your `-v` mount points
`COMFYUI_FLAGS`	`--highvram` (datacenter GPUs)	May need `--normalvram` or `--lowvram` for consumer GPUs
`RUNPOD_API_KEY`	Auto-injected	Not available (disk telemetry won't show total volume size)

Volume Mount¶

The -v mount maps a host directory to /workspace inside the container. Everything persists there:

/path/to/data/
├── studio/          ← STUDIO_DIR (app code, catalogs, jobs, config)
└── ComfyUI/         ← ComfyUI installation + downloaded models

Use a fast SSD

AI model loading speed depends on disk I/O. NVMe SSDs make a noticeable difference compared to spinning disks, especially for large models (10+ GB).

Stopping and Restarting¶

# Stop
docker stop comfyui-studio

# Start again (data persists)
docker start comfyui-studio

# View logs
docker logs -f comfyui-studio

# Remove (data in volume persists)
docker rm comfyui-studio

Updating¶

Application Updates (No Rebuild)¶

Click Check for Updates in the web UI, same as RunPod.

Docker Image Updates (Rebuild)¶

docker pull ghcr.io/diego-devita/comfyui-studio:latest
docker stop comfyui-studio
docker rm comfyui-studio
# Re-run with the same docker run command

Your data in the volume mount is preserved across image updates.

Networking¶

By default, both services bind to 0.0.0.0:

Port 8000: Studio web UI (password-protected)
Port 8188: ComfyUI graph editor (NO authentication)

For security, consider:

Only exposing port 8000: -p 8000:8000 (omit -p 8188:8188)
Binding to localhost only: -p 127.0.0.1:8000:8000
Putting a reverse proxy (Nginx, Caddy) in front with HTTPS