Synthesia built an interactive AI avatar of a TechCrunch reporter, trained on a single story

Synthesia, the British digital-avatar startup valued at $4 billion, has produced an interactive AI avatar of a TechCrunch reporter. Dominic-Madori Davis, who covers venture capital for the site, described the process in an article published on September 26: the avatar was trained on one of her stories and answers questions only about that story. According to Davis, it is the first avatar Synthesia has made for a journalist, and the first it has made for anyone outside the company apart from its own head of corporate affairs, Alexandru Voica.

Two minutes of voice, a couple of days of work

The reporter recorded her material in a small film studio inside Synthesia's new office in New York. The team took numerous photographs of her and captured a two-minute voice sample, and she had to consent to the avatars being made. Synthesia produced a personal avatar, which reads any script it is given, and interactive avatars that listen and reply, in versions with and without glasses. The work took a couple of days.

How the avatar answers

The interactive avatar combines voice-to-text, an agentic language model, text-to-voice and a video model that animates the face. The video and voice models are Synthesia's own, though customers can substitute tools from Cartesia, ElevenLabs, Google or OpenAI, and can host the result on their own cloud or pay the company to host it. The avatar is deterministic: asked personal questions that were never part of its training, it steered the conversation back to the story. Davis wrote that friends found the voice interesting and unsettling, and that her mother joked, "I don't remember giving birth to two of you."

Business beyond the demo

Synthesia sells three kinds of products. Its original platform turns written scripts into videos with classic avatars. An agentic platform called Sessions lets employees talk to avatars in surveys or roleplay sessions, where they practice sales pitches and receive scores on their answers. A third, API-based offering lets companies combine Synthesia's video and voice models with other technology. The company, which competes with D-ID, HeyGen and Colossyan, said last year that it had passed $100 million in annual recurring revenue, and it reached a $4 billion valuation earlier this year.

Trust as the open question

Davis wrote that she has mixed feelings about digital twins. She sees the appeal of cloning yourself for work and finds avatars less jarring than humanoid robots, but doubts that trust, the core of journalism, can be handed to software. Because her avatar is deterministic, it never improvises; she noted that a non-deterministic version, free to speak at length, could invite unhealthy attachment. One investor she spoke to rejected the idea of avatar news anchors outright, while others were unsure.