Replicate

Replicate runs open machine-learning models — image, video, audio and language — as hosted jobs you start over an API and collect when they finish.

Authentication

• API token

Actions

Generic API request. Call any Replicate API endpoint on this connection, for the case where no shipped action covers it yet. The connection supplies the base URL, the authentication and the egress route, so only the path, query and body are yours to fill.

• Cancel predictionDestructiveStop a prediction that is still running - when a long generation is no longer wanted, or the caller gave up - so it stops consuming compute.
• Create predictionRun a model on Replicate by naming a model version hash and passing that version's inputs.
• Get modelRetrieve a single model by owner and name (GET /models/{modelowner}/{modelname}).
• Get predictionPoll a prediction started with Create prediction until it settles, then read its result.
• List modelsBrowse the public models available on Replicate, to resolve an owner/name pair before creating a prediction.
• List predictionsList the predictions created by the account, most recent first, to poll or audit recent runs and recover their output URLs.

Triggers

• New or updated public modelStarts when a public model is published or gets a new version on Replicate.
• New predictionStarts when a new prediction is created in the connected Replicate account.
• Prediction completed (webhook)Starts the moment a Replicate prediction reaches a terminal state (succeeded, failed or canceled), delivered by a webhook instead of polling the prediction.

Product names, logos and brands are the property of their respective owners and are used here for identification only. Their use does not imply endorsement by, or affiliation with, those owners.