TechCrunch journalist gets her own talking AI avatar from Synthesia

Synthesia, a video-generation startup valued at $4 billion, has created an interactive digital avatar of a TechCrunch journalist — the first such avatar the company has built for a journalist or anyone outside its own staff.

The avatar was made at Synthesia's new New York office, where the journalist was photographed and recorded a two-minute voice sample. The team produced both simple avatars that read a fixed script and interactive versions that can listen and respond. The interactive avatar was trained only to answer questions about a specific article she wrote on fraud at venture-backed startups.

How the technology works

The avatar's stack combines a voice-to-text model, an agentic language model that interprets and responds to input, a text-to-voice model, and a video model built by Synthesia that animates the avatar while it talks. Clients can also substitute voice and video models from providers like Cartesia, ElevenLabs, Google, or OpenAI. Building the avatars took Synthesia's team a couple of days.

Because the avatar is deterministic, it only responds within its trained scope and redirects any off-topic questions back to the article it was trained on. When the journalist's family and friends tested it, they described it as impressive but somewhat unsettling, noting the voice and likeness weren't perfectly accurate.

Synthesia builds three types of products: a video-creation platform with standard avatars, an agentic platform called Sessions for roleplay and surveys, and an API platform for building custom interactive avatars. The experience prompted reflection on whether avatars could eventually augment or replace journalists, though the piece notes that trust remains a distinctly human element that seems difficult to automate.