Back to projects

Immortalisation — How to Build an AI Magical Portrait with Voice Cloning & Personality AI

A technical walkthrough of building an AI-powered magical portrait: ElevenLabs voice cloning, RAG personality modelling, D-ID lip-sync animation, and real-time WebRTC streaming.

Layer 1: Legacy capture

Voice samples (30-60 min for ElevenLabs PVC), written personality (WhatsApp exports, emails in a RAG knowledge base), movement and expressions (webcam footage for digital twin training).

Layer 2: The personality model

RAG knowledge base with vector embeddings. Long-term memory with Mem0 or Zep. Behavioural guardrails via system prompt engineering. Tone calibration from written corpus.

Layer 3: The portrait

Still portrait with voice, light animation with eye movement and head tilts, or full real-time avatar with lip-sync via D-ID Streaming API.

Layer 4: Ethics & consent

Explicit pre-mortem consent required. Anti-dependency design. Family-controlled governance with kill switch. AES-256 encryption at rest.