Replicate Automation

Replicate Automation - AI Model Workflow Automation

Skills Overview

Replicate Automation integrates with Composio MCP to automatically run Replicate AI model predictions, upload input files, check model architecture, list versions, and manage prediction history.

Use Cases

Image Generation Automation: Batch-run image models like Flux and SDXL, automatically handle prompts and output files

LLM Inference Workflows: Automate calls to large language models such as Llama and GPT, supporting both synchronous and asynchronous inference

Model Version Management: Check model updates, view version history, and fetch the latest model architecture and documentation

Core Features

Run Model Predictions: Supports synchronous/asynchronous predictions for any public model; you can set wait time and webhook callbacks

File Management: Upload binary files such as images and documents as model inputs; supports file lists and history queries

Fetch Model Information: One-click retrieval of model input/output architecture, version lists, and README documentation

FAQs

How do I run Replicate AI model predictions?

Use the REPLICATE_MODELS_PREDICTIONS_CREATE tool, providing the model owner (e.g., black-forest-labs) and model name (e.g., flux-1.1-pro), along with model-specific input parameters. For quick tasks, set the wait_for parameter to synchronously wait for results; for long-running tasks, use webhooks to receive asynchronous notifications.

What’s the difference between synchronous and asynchronous predictions?

Synchronous predictions set the wait_for parameter (1–60 seconds) to directly wait for results, suitable for fast models. Asynchronous predictions do not set this parameter; the task runs in the background, and once completed, you retrieve results via webhook or by polling REPLICATE_PREDICTIONS_LIST. This is ideal for long-running image generation or video processing tasks.

How do I view a model’s input/output schema?

Use the REPLICATE_MODELS_GET tool by passing the model owner and name. The openapi_schema field in the result contains the complete input/output definitions. Input key names and types vary greatly across different models, so it’s recommended to check the schema before each prediction run to avoid parameter errors.