​
Workflows

​
Description

A workflow is a ComfyUI / ComfyStream concept representing the blueprint of an AI task, often as the combination of multiple nodes. Data flow is managed through a directed acyclic graph (DAG); it is unidirectional but may incorporate multiple inputs or execution chains.

​
How it’s built

Workflows are most commonly implemented through the ComfyUI interface and developed+tested using a development environment including ComfyStream, ComfyUI, and a GPU provider like RunPod or a local GPU.

​
Pipelines

​
Description

A Livepeer pipeline is a workflow that is deployed on the Livepeer network and is available for public or private usage by streamers or application developers. It constitutes a unidirectional data-in/data-out flow, the behavior of which may be changed through workflow parameter updates. Right now, pipelines are simply deployed ComfyUI Workflows; in the future, a pipeline may combine multiple inference engines (for example, ComfyStream + Blender)

​
How it’s created

Pipelines are deployed via the Livepeer Pipelines UI.

Customize SD LoRas