Skip to main content
Image training has no audio, so you attach a voice on the same Create Face request. Send default_voice_id or voice_name, not both. See Choose a Model and Training Path for photo vs video, and Voices for the full Voices API.

​
Option 1. Train your own voice

Create a Tavus Voice from a sample of audio in PAL Maker or with Create Voice. When it is ready, pass its id as default_voice_id.
Use this when you want the face to sound like a specific person. You must hold the rights to that voice. See Platform Policies.

​
Option 2. Use a stock voice

This is the faster path if you just want the face up and running. Pick a stock voice from the catalog and attach it now. You can change the face’s default later with Update Face. Two ways to point at a stock voice:
You are not stuck with that voice in every conversation. A PAL can still set its own voice_id or external_voice_id. See Text-to-Speech (TTS).

​
Example stock voices

These are example voice_name slugs, each with a short sample clip.