TechCrunch journalist Joanna Stern documented her experience creating a personalized, interactive AI avatar of herself at Synthesia's New York office. Synthesia, a UK-based startup that recently hit a $4 billion valuation, specializes in AI-driven video avatars and interactive training tools.
The process involved Stern visiting Synthesia's office, where numerous photos were taken and a two-minute voice recording was captured. The resulting interactive avatar is trained on Stern's specific article about venture-backed startups and fraud, meaning it can only answer questions related to that topic. This contrasts with 'personal' avatars that can read any script.
The technology powering these avatars combines voice-to-text, agentic language models, text-to-voice, and Synthesia's own video models, with options to integrate third-party AI services like those from ElevenLabs or OpenAI. Synthesia offers three main product types: a platform for creating scripted avatar videos, an interactive 'Sessions' platform for role-playing and surveys, and an API for custom avatar development.
Stern noted that while the personal avatar's voice was accurate, friends found the interactive avatar's likeness and voice less convincing but still 'creepy.' Her parents also tested the avatar, with her mother joking about 'giving birth to two' of her. The experience prompted Stern to consider the future of journalism and the role of AI avatars in presenting news.