InterVideo – Live AI Avatar Conversation App

Talk live with an AI avatar animated from a single photo. Your webcam and the avatar feed display side by side. Sessions record to local video files. Choose any persona via system prompt (interviewer, storyteller, teacher, etc.). Conversation memory persists across sessions.

Features

• Live avatar animation from a single photo using MediaPipe face landmarks
• Real-time conversation with AI (speech-to-text → LLM → text-to-speech with viseme sync)
• Local recording to WebM format (480p, audio + video mixed)
• Persistent memory via JSONL transcripts and rolling summaries
• No local AI models – runs on any modern laptop from ~2020+

Tech Stack

Backend: Node.js 26.x + TypeScript
Frontend: Vite + vanilla TypeScript + Canvas 2D
STT: Deepgram Flux
LLM: OpenAI GPT-5.4-nano (Responses API)
TTS + Visemes: Azure Cognitive Services Speech
Face Animation: MediaPipe Face Landmarker

Quick Start

1. Install dependencies

   npm install

2. Set up API keys in .env

   DEEPGRAM_API_KEY=<your-key>
   OPENAI_API_KEY=<your-key>
   AZURE_SPEECH_KEY=<your-key>
   AZURE_SPEECH_REGION=<region>
   

3. Start the dev server

   npm run dev

4. Open http://localhost:5173 in Chrome or Edge

Cost

$0.016 per conversation minute ($0.95/hour). Deepgram new accounts get $200 credit.

Requirements

• Chrome or Edge (current version)
• 2+ CPU cores, 8GB RAM
• ~0.5–1 GB disk per hour of recording
• API keys for Deepgram, OpenAI, and Azure Speech

Deployment

See the brief in /meta/brief.md for full implementation details, pipeline architecture, and constraints.

---

Expect ~1–3s turn latency. Use only photos you own or have consent for.