Replicate Action Block

What it does: Runs open-source models hosted on Replicate (opens in a new tab) — image, video, audio, and language models — and brings their output back into your workflow.

🎨

In simple terms: Pick any model from Replicate (FLUX for images, Whisper for transcription, a Llama variant for text…), give it a JSON input, and get back the generated file URLs or text.

Credentials

Create an API token at replicate.com/account/api-tokens (opens in a new tab) and paste it when connecting the block.

Actions

Action	What it does	Output
Run Model	Runs a model by `owner/name` or `owner/name:version`	Prediction object + `output` + `urls`
Run Deployment	Runs one of your deployments (`owner/name`)	Prediction object
Get Prediction	Fetches a prediction, optionally waiting for it to finish	Prediction object
Cancel Prediction	Cancels a running prediction	Prediction object
List Predictions	Your recent predictions	Paginated list
Get Model	Model metadata and latest version	Model object
List Model Versions	All versions of a model	Version list
Search Models	Finds public models by search term	Paginated model list

Waiting for results

Model runs are asynchronous by default. The Wait (seconds) field (0–60) sends Replicate's Prefer: wait=<n> header, which holds the request open until the prediction finishes:

Wait set (e.g. 60) — the block returns the completed prediction, with output already populated. Best for fast models.
Wait empty — the block returns immediately with id and status: "starting". Follow up with a Wait block and Get Prediction using {{replicate.id}}. Best for slow video/large-image models.

Output shape

The block returns the full prediction object, plus two conveniences:

output — the model's raw output (a string, an array, or an object depending on the model).
urls — present only when the output is a URL or an array of URLs, so you can pipe generated images/videos straight into a send-message or upload block.

Tips

Model accepts owner/name (always the latest version) or owner/name:versionHash (pinned — recommended for production, since a model update can change the output shape).
Input is the exact JSON shown on the model's API tab on replicate.com, e.g. {"prompt":"a cat astronaut","aspect_ratio":"16:9"}.
Set a Webhook URL with a Webhook Events Filter of completed to be notified when a long run finishes instead of polling.
Predictions bill per run — pin a version and cap waitSeconds so a stuck run doesn't hold your workflow open.

Razorpay Resend