replicate

name: replicate description: "Replicate - run 1000+ AI models (FLUX, Stable Diffusion, Whisper, LLaMA, etc.) via API. Use when user needs a specific AI model not available natively."

Replicate API Skill

Overview

Run 1000+ open-source AI models via API. Image generation (FLUX, SDXL), video, audio, text, and more.

API Key

import os
REPLICATE_API_KEY = os.getenv('REPLICATE_API_KEY')
# Key from ~/.claude/.credentials.master.env

MCP Server

Already configured in ~/.claude/mcp.json as replicate. Can use MCP tools directly.

Dependencies

pip install replicate

Basic Usage

import replicate

# Image generation with FLUX
output = replicate.run(
    "black-forest-labs/flux-1.1-pro",
    input={
        "prompt": "a beautiful sunset over mountains, photorealistic",
        "aspect_ratio": "16:9",
        "output_format": "png"
    }
)

# Stable Diffusion XL
output = replicate.run(
    "stability-ai/sdxl:39ed52f2a78e934b3ba6e2a89f5b1c712de7dfea535525255b1aa35c5565e08b",
    input={
        "prompt": "an astronaut riding a horse on mars",
        "negative_prompt": "blurry, low quality",
        "width": 1024,
        "height": 1024
    }
)

Popular Models

Async Predictions

prediction = replicate.predictions.create(
    model="black-forest-labs/flux-1.1-pro",
    input={"prompt": "..."}
)
prediction = replicate.predictions.get(prediction.id)
print(prediction.status)  # "starting", "processing", "succeeded", "failed"
print(prediction.output)  # URL when done

Tips

1. Check model page on replicate.com for input parameters
2. Use replicate.models.search("keyword") to find models
3. Output is usually a URL - download with requests
4. Billing is per-second of GPU time