Replicate

Replicate lets developers run machine learning models through a cloud API without managing servers. It hosts a large catalog of models for images, audio, video and text.

You pick a model, send an input such as a prompt or an image, and receive the output from a single call. Client code can be written in Node, Python or plain HTTP. You can also upload and deploy your own custom model, and fine-tune supported models on your own data.

• Run hosted models with a few lines of code
• Image, speech, music and video generation
• Image restoration and captioning models
• Deploy your own custom models
• Fine-tune supported models

• App developers adding generative media features
• Teams comparing many models before they choose one
• Researchers who want to deploy their own models

The service has a free way to start, and paid usage is charged for the compute each model uses. Costs depend on the model and hardware, so compare the model pages before you scale. Models that have not been used recently can take longer to respond on the first call while they start up.