Realistic Avatars
Choose Realistic Avatars when you already have a voice agent, or a text-to-speech voice, and want a photorealistic person to speak it. You keep full control of the conversation: your server streams audio to the avatar, and your page (or your LiveKit room) shows it.- Works with any TTS or speech-to-speech model.
- Interruptions are one call, and tell you how much of the reply was heard.
- Your API key stays on your server.
Voice Agents
Choose Voice Agents when you want a talking AI agent without building one. You describe the agent in plain language, pick an avatar and a voice, and share it. The realtime AI model, speech recognition and voice are all included in the per-minute price.- Templates for support, sales, interviews, reception and tutoring.
- Tools that call your HTTPS endpoints (or a Zapier, Make or n8n webhook) during the call.
- A transcript of every call, and a signed post-call webhook.
- Callers talk from their web browser. Phone calls aren’t supported.
Open Source
Choose Open Source when you have (or want) your own 3D avatars and prefer something free that runs entirely in the browser. Audio never leaves the page.- A 1 MB lip-sync model that runs as plain JavaScript.
- Works with any audio: a TTS clip, a WebRTC voice agent, or a microphone.
- A framework-free core and a React component.
All three ship in one npm package,
@interviewflowai/talking-avatar. Realistic Avatars use its /server and /live imports. Open Source uses the root import and /react.