Realtime Avatar turns a portrait or source video into a talking avatar. Drive it two ways:

Realtime turns

Stream audio-clocked talking-avatar video as the model speaks — for live conversation and co-hosting.

Non-streaming lipsync

Pass a pre-recorded audio URL, get back a rendered MP4 URL. Ideal for backend and batch jobs.

​
Start here

Quickstart

A talking avatar in a few lines, in TypeScript or Python.

Authentication

API keys, environments, and keeping live keys out of the browser.

TypeScript SDK

realtime-avatar — client, streaming player, React hooks.

Python SDK

realtime-avatar — lean httpx client for backend and batch.

​
How it works

1

Create an avatar

Register a portrait image or a source video as an avatar (ava_…).
2

Prepare

Warm the avatar once so the first turn is fast. (Lipsync prepares implicitly.)
3

Drive it

Either stream a turn (live video) or lipsync an audio URL (MP4 URL back).
Machine-readable references for agents and tooling: the OpenAPI spec, llms.txt, and a remote MCP server at /api/mcp.