Inside Synthesia’s Interactive AI Twin and How It Works
Synthesia creates a journalist's interactive AI twin, showing how avatars work, their business uses, and why trust remains the central unresolved risk.
Summary
On September 26, 2026, journalist Dominic-Madori Davis unveiled four AI avatars that Synthesia created from numerous photographs and a two-minute voice recording in its new New York office. The process took a couple of days and produced personal and interactive versions, each with and without glasses. Personal avatars read supplied scripts. The deterministic interactive twins answer only questions about Davis’s story on why venture-backed startups commit more fraud than non-VC-backed companies, redirecting personal questions to that subject. Synthesia had previously made such an avatar only for corporate affairs chief Alexandru Voica, whose twin handles common press questions.
Originally based in the U.K., Synthesia competes with D-ID, HeyGen and Colossyan. It reached a $4 billion valuation in 2026 after exceeding $100 million in annual recurring revenue in 2025. Its products include a video creation and distribution platform for scripted avatars, the agentic Sessions platform for interactive surveys and roleplay, Roleplay Sessions for scoring employee practice, and APIs combining Synthesia models with other services. Interactive avatars use voice-to-text, an agentic language model, text-to-voice and Synthesia’s video animation. Customers may substitute technology from Cartesia, ElevenLabs, Google or OpenAI and use their preferred cloud or Synthesia hosting.
Davis found the personal avatar’s voice accurate, but friends considered both versions interesting and creepy and judged the interactive twin’s voice and likeness less convincing. Her mother called it amazing. The experiment highlights potential always-available training, communications and workplace support, but raises questions about AI-generated news, replacing journalists, trust and psychological harm from nondeterministic twins. Davis expects corporate adoption to evolve but doubts Gen Z will normalize digital doubles, while finding avatars less disturbing than humanoid robots because users can log off.
Positives
- $4 billion was Synthesia’s 2026 valuation after annual recurring revenue exceeded $100 million in 2025.
- Four avatars were created from photographs and two minutes of recorded speech within a couple of days.
- Roleplay Sessions lets employees practice activities such as sales pitches and receive scores from an interactive avatar.
- Cartesia, ElevenLabs, Google and OpenAI models can replace parts of Synthesia’s stack, giving enterprise customers technical flexibility.
- Deterministic controls kept Davis’s interactive twin focused on its assigned article and prevented it from answering personal questions.
Risks & concerns
- Friends described the avatars as creepy and found the interactive version’s voice and likeness less convincing than the personal avatar.
- Journalistic trust may not transfer to an AI presenter, limiting avatars’ ability to augment or replace human reporters.
- AI-generated material already faces pushback on social media and news-sharing platforms.
- Nondeterministic digital twins could encourage psychological distress by appearing to think, improvise or possess awareness.
- Davis expects Gen Z may never become comfortable with digital doubles despite their potential workplace utility.