How do I connect Replicate to Lodol?

Authenticate Replicate once from your workspace and every action and trigger it supports becomes available to drop into a workflow. There is no glue code to write or maintain.

All integrations

Replicate

AI ModelsImages & Design

Run any of the thousands of AI models on Replicate (image, video, audio, speech and language) and collect their output. Also covers the account behind them: models and versions, deployments, fine-tuning runs it is watching, uploaded files and collections.

Get Started Browse all integrations

31 actions

Actions

Steps your workflow can run in Replicate.

Get the connected accountLook up the user or organization this Replicate connection belongs to. Handy for filling in the owner of a model you are about to create, so nothing is hard-coded.

Get the webhook signing secretFetch the key Replicate signs its webhook requests with, so a workflow that receives them can prove they really came from Replicate.

Search ReplicateSearch Replicate for models, collections and documentation pages in one go. The best way to find a model when you know what you want it to do but not what it is called.

Search public modelsSearch only the public model catalogue, a page at a time. Use 'Search Replicate' instead if you also want collections and documentation.

List public modelsBrowse Replicate's public models, newest release first by default. Read the next page by passing back the next_cursor this returns.

List model collectionsList the collections Replicate curates: 'Text to image', 'Super resolution' and the rest. Each one's slug opens it with 'Get a collection'.

Get a collectionOpen one curated collection and get every model in it, each with the input schema its predictions take.

List available hardwareList the hardware a model or deployment can be set to run on, with the SKU each one is named by.

Get a modelLook up one model, including the input schema its predictions take and the version it currently resolves to. Read input_schema to find out what to put in a run's Model input.

Create a modelCreate a new model on Replicate. It starts empty: a home for versions pushed with Cog, or for the weights a training produces. The owner has to be the account this connection belongs to, or one of its organizations.

Update a modelChange a model's description, README or the links on its page. Fill in only what you want changed; anything left blank is left as it is.

Delete a modelDelete a model you own. Replicate only deletes private models with no versions left, so a model that has had one pushed to it has to have its versions removed on replicate.com first.

Get a model's READMERead a model's README: the prose on its page, in Markdown. Usually where the author explains what the inputs mean.

List a model's examplesList the example predictions a model's author saved. Each one shows a real input and the output it produced, which is the quickest way to see what to send the model.

List a model's versionsList a model's versions, newest first, each with the input schema it takes. Use this to pin a run to a version that will not change under you.

Get a model versionLook up one version of a model and the exact input and output schemas it declares.

Run a modelRun any model on Replicate. Set 'Wait up to' and the finished output comes back in this step; leave it blank and the run goes on in the background, to be picked up later with 'Get a prediction'. Leave the version blank to run whichever version the model is on now. For one of Replicate's own official models, 'Run an official model' is the simpler step.

Run an official modelRun one of the models Replicate maintains itself, such as black-forest-labs/flux-schnell. Official models are always on, so they are run by name and always use the current version. For any other model use 'Run a model'.

Run a deploymentSend a prediction to one of your deployments, so it runs on the hardware and version that deployment is configured for.

Get a predictionCheck on a run and collect its output. Use this after starting a run in the background: succeeded tells you the output is ready, and output_files holds any images, audio or video it produced.

List predictionsList the account's runs, most recent first: from the API and from the Replicate website. Filter by when they were created to look at one day's activity. Replicate keeps this listing to roughly the last three months.

Cancel a predictionStop a run that is still going, so it stops costing anything. A run that has already finished is left as it is.

List trainingsList the fine-tuning runs on this Replicate account, most recent first, with what each one produced. Trainings are started on replicate.com; this watches them.

Create a deploymentCreate a deployment: a named endpoint that always runs one model version on hardware you choose. Leave the minimum at 0 and nothing runs (and nothing is billed) until a prediction arrives.

Get a deploymentLook up one deployment: the model version it serves, the hardware it runs on and how far it may scale.

List deploymentsList the account's deployments and what each one is currently serving.

Update a deploymentRescale a deployment or move it onto another version or hardware. Fill in only what you want changed. Every change publishes a new release.

Upload a fileUpload a file to Replicate and get back a URL you can hand to a model that takes a file input. Better than inlining a large file into every run: upload it once and reuse the URL. Replicate keeps the file for a limited time and reports when in expires_at.

List filesList the files this account has uploaded, most recent first, with when each one expires.

Get a fileLook up one uploaded file: its size, checksum, metadata and when Replicate will delete it.

Delete a fileDelete an uploaded file. Anything still pointing at its URL stops working, so delete it once the runs that used it are done.