This is not Pixio Integrations. That API gives you 500+ curated models with fixed inputs. This runs your ComfyUI graph. Different product, different API, different key.
Which one do I want?
If a catalog model does the job, use the app — it’s faster and there’s nothing to maintain. Reach for ComfyUI when you need control the catalog can’t give you.
Base URL and auth
Endpoints
How it works
1
Build a workflow
Author your graph in ComfyUI, exposing inputs with Pixio’s external input nodes — see workflow inputs.
2
Deploy it
A deployment makes a specific workflow version callable. It gets a
deployment_id, which is what your code references.3
Queue a run
POST /run/deployment/queue with the deployment_id and your inputs. You get a run_id back immediately — execution is asynchronous.4
Get the outputs
Pass a
webhook when queueing and receive status pushes including the terminal one, or poll GET /run/{run_id}. Webhooks are preferred.Core concepts
The distinction that matters: you deploy a version and you call a deployment. Editing your graph changes nothing about your API calls until you deploy — which is what makes production safe.
Billing and concurrency
- Billing posts at the terminal state, for the time actually used
- Queue time is unbilled
- Your plan’s Concurrent GPUs limit decides how many run at once; runs beyond it wait in
queuedrather than failing
That last point is why a burst of queued runs isn’t an error. They’re waiting for a GPU, not broken.
Where to next
Getting started
Queue your first run and fetch its outputs.
Run lifecycle
All nine statuses, webhooks vs polling, cancelling.
Workflow inputs
Exposing inputs with external nodes, including images.
API keys
Creating, scoping, and revoking keys.