Replicate
by Replicate, Inc.
Integrate an AI agent with Replicate to run open-source models for image generation, video, audio, and language tasks — the agent picks a model from the library, starts a prediction, and collects the result by polling or webhook when the run finishes.
How agents use Replicate
- ✓Agent generates images using SDXL or Flux based on a text prompt from a workflow
- ✓Agent runs an upscaling model to improve image quality before delivering to a client
- ✓Agent processes audio with a Replicate model for noise removal, separation, or enhancement
- ✓Agent accesses specialized fine-tuned models not available through other providers
- ✓Agent generates product images or visual content as part of an e-commerce or marketing workflow
- ✓Agent searches the Replicate model library and reads a model input schema before building a request, rather than depending on a hardcoded model ID
- ✓Agent starts a long-running prediction and either polls its status or takes a webhook on completion, so a slow model does not block the rest of the workflow
- ✓Agent cancels a prediction that is no longer needed, so an abandoned run stops consuming compute
Agent actions
Inputs: model, version, input, webhookUrl
Returns: predictionId, status, output, outputUrls
Inputs: predictionId
Returns: predictionId, status, output, outputUrls, error, metrics, prediction
Inputs: predictionId
Returns: success, prediction
Inputs: query
Returns: models, modelCount, nextCursor
Inputs: owner, name
Returns: modelId, description, latestVersion, inputSchema, model
Example workflows
Image generation pipeline
Agent takes a text prompt from a user input node, runs SDXL on Replicate, and uploads the result to Google Drive.
Model discovery before a run
Agent searches the Replicate model library for one that fits the task, reads its input schema, and builds a valid request instead of relying on a hardcoded model ID.
Long-running prediction with a webhook
Agent starts a video or upscaling prediction, registers a webhook for completion, and carries on with the workflow rather than blocking until the model finishes.
Audio cleanup before transcription
Agent runs a noise-removal or speaker-separation model over a recording, then passes the cleaned audio to the next step in the workflow.
About Replicate
- Vendor
- Replicate, Inc.
- Pricing Always review details with the vendor
- Paid — Pay per prediction — billed by CPU/GPU seconds. Image generation (SDXL): ~$0.002–0.01/image. No free tier, but new accounts receive $5 in credits.
- Authentication
- API key
- Rate limit Always review details with the vendor
- 60 requests / minute
- Compatible nodes
- AgentResourceOutput
- Website
- https://replicate.com
Put Replicate behind your AI agent
Set up Replicate alongside your other tools in the planner — no code — so AI can act on it for your customers.
Open the planner