Skip to main content

​
General

Talking Avatar, by InterviewFlowAI, gives AI voice agents a talking face. Realistic Avatars are photorealistic avatars rendered on our servers and streamed to your page, for $0.02 a minute. **Voice Agents** are complete no-code AI agents on those avatars, for $0.06 a minute. Open Source is a free, MIT-licensed 3D talking avatar that runs your own .glb model in the browser.
Realistic Avatars when you have your own voice agent and want a photorealistic person with no 3D work. Voice Agents when you want a complete agent with no code. Open Source when you have, or want, your own 3D avatars and prefer it free and fully in the browser. See Choose a product.
No. Both avatar products follow the audio alone, so they work with any provider and any voice.
Any. OpenAI, ElevenLabs, Azure, Fish Audio, Cartesia or any other TTS, and voice agents built on OpenAI Realtime, LiveKit Agents, Pipecat or your own stack.
Yes, all of it. Realistic Avatars and Voice Agents include commercial use. Open Source is MIT licensed and free for commercial and noncommercial use.

​
Pricing

Realistic Avatars are $0.02 a minute and Voice Agents are $0.06 a minute, both billed per second from prepaid credits. Open Source is free. There’s no subscription, setup fee or minimum. See Pricing.
Yes. New accounts get $1 of free credits: 50 minutes of Realistic Avatars, or about 16 minutes of Voice Agent calls. You can also talk to a realistic avatar on the live demo without signing up, and try every avatar for free from the dashboard’s Avatars page.
The price is the same at any volume. For very large volumes, email support@talking-avatar.dev.
Set a monthly spending limit. Once it’s reached, sessions and calls stop until next month.

​
Realistic Avatars

A growing library of ready-made photorealistic avatars, women and men of many ages and backgrounds, in professional and casual looks. Every account can use all of them. See Avatars.
Yes. A custom avatar is $199 per avatar, one time, with the signed consent of the person on camera.
Yes. The avatar can join your LiveKit room as a participant, with a publish-only token you mint. See LiveKit.
Yes. clear() stops the avatar mid-sentence and tells you how much of the reply the user heard.
Up to 3 hours each. Run as many sessions as you need.
No. The SDK’s server part runs on Node.js (22+, or 18/20 with ws), but any language can use the realtime avatar API directly, over one WebSocket. See the Python example.
Any. The avatar follows the audio, whatever the language.

​
Voice Agents

No. Describe the agent in plain language, pick a face and a voice, and share its link. Adding it to a website is one line to paste. Tools and the webhook are optional, and can point at Zapier, Make or n8n.
In their web browser, on desktop or mobile, with their microphone: from the agent’s link, a button on your website, or an embedded frame. Phone calls and phone numbers aren’t supported.
It understands callers in their language. To have it reply in their language, pick a Multilingual voice and keep Reply in the language they speak in the prompt. English voices speak English only.
Yes. Every call is on the agent’s Calls tab with its transcript, and the post-call webhook can send it to your systems.
Yes. Make it private: only people with one of its expiring links can call.
Yes. Remove watermark is $19 a month and removes the watermark and the Powered by Talking Avatar line.

​
Open Source

No. Everything runs in the browser. The only network requests are for the avatar and the model file.
Any .glb with the Oculus viseme morph targets, or other morph names mapped with morphTargets. See 3D avatars.
The model is trained on English and holds up reasonably on other languages tested, using the nearest English mouth shape for sounds English lacks. See How it works.