Skip to main content
Before your application can run workflows, ComfyUI has to be running somewhere. There are three ways to get a live deployment, and they differ mainly in who manages the GPUs and how the deployment scales.

Serverless API

Deploy a versioned ComfyUI environment as a managed endpoint with on-demand GPU capacity.

Comfy Cloud

A managed, always-available ComfyUI. No environment to build, models are pre-installed.

Self-Host

Run ComfyUI on your own hardware, a rented GPU, or a container. Full control.

​
Which deployment should I use?

All three run the same workflows. A workflow you build and test on a self-hosted instance can move to Serverless or Cloud without changes, because every surface accepts the same API format.

​
Choosing in practice

  • You want a production endpoint for a specific workflow. Use the Serverless API. You define the environment once, cut a release, and get a URL with managed GPU capacity behind it.
  • You want to skip infrastructure entirely. Use Comfy Cloud. It is effectively a deployment that is always there: models are pre-installed and the endpoint never goes away.
  • You already have GPUs, or need hardware and models under your control. Self-host, on your own machine or on a GPU cloud such as Runpod or Vast.ai.

​
Next step: run workflows against it

Once a deployment is live, see Running Workflows to call it from your application. The Comfy SDKs work against all three targets; self-hosted instances additionally need the API Proxy during the v2 beta.