Talking Avatar
Realistic AI avatars that give your voice agent a face
Start new thread

Talking Avatar - Realistic AI avatars that give your voice agent a face

by•
Give your AI voice agent a lifelike face. Talking Avatar delivers real-time lip-sync, integrates with OpenAI, ElevenLabs, Cartesia, and other TTS providers via WebRTC or LiveKit, and also lets you build voice agents without code. Includes a free, open-source, MIT-licensed 3D avatar that runs right in your browser.

Add a comment

Replies

Best
Maker
📌

Hey Product Hunt 👋

I'm Tushar, founder of Talking Avatar and CTO at InterviewFlowAI. We build AI interviewers, and our voice agents kept hitting the same wall: people trust a face more than a voice, but real-time avatars were too expensive and too slow for long conversations. So we built our own, and today we're opening it up.

Talking Avatar gives any voice agent a realistic face, three ways:

  • Realistic Avatars API. Stream your agent's TTS audio to a photorealistic avatar with real-time lip sync. One npm package; works with OpenAI, ElevenLabs, Cartesia, LiveKit, Pipecat or plain WebRTC.

  • Voice Agents, no code, all-in. Avatar, LLM, speech recognition and voice in one price. Write a prompt, pick a face, then share a link or embed it on your site.

  • Open Source, free (MIT). A 3D talking avatar SDK that runs entirely in the browser with a 1 MB model. No server, no API key:

Billed per second, no subscription, and every new account gets $1 in free credits (about 50 avatar minutes), no card needed.

🎁 Product Hunt offer: use code PRODUCTHUNT on the Billing page to double your first top-up (up to $100 extra), until Dec 4.

We'd love your feedback on:

  1. Which avatar feels most natural to you? You can talk to one live on the homepage.

  2. What would you give a face to: support, tutoring, sales, interviews, something else?

Mukul and I are here all day and will answer every question 🙏

Maker

I’m Mukul, one of the makers of Talking Avatar. We trained the photorealistic avatar model on paired video and audio using NVIDIA GPUs, then optimized inference so it can animate the face as your agent’s audio streams in over WebRTC or LiveKit.

The open-source 3D SDK takes a different approach. It uses a 1 MB model to predict 22 mouth shapes every 10 ms, entirely in the browser.


If you open the live demo, ask an unexpected follow-up and see how well the mouth movements track the reply. Would love for you to try it out and give your feedback.