Replicate

Featured models

laion-ai/erlich

Generate a logo using text.

143,595 runs

kuprel/min-dalle

Fast, minimal port of DALL·E Mini to PyTorch

422,658 runs

nightmareai/disco-diffusion

Generate images using a variety of techniques - Powered by Discoart

45,796 runs

nightmareai/cogvideo

Text-to-video generation

6,875 runs

pixray/text2image

Uses pixray to generate an image from text prompt

524,738 runs

jingyunliang/swinir

Image Restoration Using Swin Transformer

148,371 runs

Publish models

Define your model with Cog, an open-source tool to package a machine learning model in a container:

build:
  gpu: true
  system_packages:
    - "libgl1-mesa-glx"
    - "libglib2.0-0"
  python_version: "3.8"
  python_packages:
    - "torch==1.8.1"
predict: "predict.py:Predictor"

Then, push it to Replicate:

$ cog push

Your model can now be run in the cloud on the web and via an API.

Replicate is not open for everyone to create models yet. Join our waitlist and we'll notify you when we're ready for new users.

Run models from your code

Run open-source or private models with just a few lines of Python code:

import replicate
model = replicate.models.get("hooli/hotdog-detector")
model.predict(image=open("possible-hot-dog.jpg"))

Or query the API directly using your tools of choice:

curl -s -X POST \
-d '{"version": "...", "input": {"image": "https://example.com/image.jpg"}}' \
-H "Authorization: Token $REPLICATE_API_TOKEN" \
-H 'Content-Type: application/json' \
https://api.replicate.com/v1/predictions

Explore models or, learn more about our API

Machine learning doesn’t need to be so hard.

Run open-source models with a cloud API.