Synthesia builds interactive AI avatar of journalist, sparking debate on AI in media

Synthesia, the London-originated video-generation company, recently delivered a working interactive avatar of TechCrunch journalist Dominic Madori Davis. The digital twin can converse only about a single pre-selected story, a demonstration that underscores how far AI-driven avatars have progressed and raises questions about their role in public relations and journalism.

Founded in 2019, Synthesia has positioned itself among fast-growing avatar firms such as D-ID, HeyGen and Colossyan. The startup announced a $4 billion valuation earlier this year and reported crossing $100 million in annual recurring revenue in 2023. Its core offerings include a platform for creating scripted videos with AI avatars, an agentic “Sessions” product that lets employees role-play scenarios with responsive avatars, and an API that lets developers combine Synthesia’s video and voice models with external services.

For the journalist’s avatar, Synthesia set up a mini studio inside its new New York office, captured dozens of photographs and recorded a two-minute voice sample. The company then generated four versions: two “personal” avatars that read any supplied script, and two “interactive” avatars that listen, process speech, and reply. The interactive models rely on a pipeline that converts spoken input to text, feeds the text to a language model that determines the appropriate response, synthesizes the reply into audio, and finally animates the avatar with Synthesia’s proprietary video model. Customers may swap out any component with alternatives from providers such as Cartesia, ElevenLabs, Google or OpenAI, and they can host the resulting avatars on their own cloud infrastructure or on Synthesia’s servers.

During the trial, the journalist tested the personal avatars by having them speak about autumn in New York, noting that the synthetic voice captured his tone without the hoarseness present in the original recording. The interactive avatars, however, were deliberately deterministic: they were trained solely on his article about venture-backed startup fraud and redirected every query back to that narrative, even when asked personal questions. Friends who viewed the demo described the experience as both fascinating and unsettling, while his parents attempted to stump the avatar with family-specific queries, only to receive the same story-centric answer.

The experiment has ignited a broader conversation about the future of media. Some investors expressed skepticism that audiences would accept news delivered by a virtual counterpart, citing concerns over authenticity and the recent backlash against low-quality AI content. Others suggested that avatars could augment reporters by handling routine inquiries or providing multilingual versions of stories. For corporate communications, the technology offers a scalable way to field common press questions without tying up human staff, yet the deterministic design also highlights the limits of current models in handling nuanced, off-script interactions.

As AI avatars become more accessible, companies will need to balance novelty with trust. Synthesia’s demonstration shows that creating a convincing digital twin can be done in a matter of days, but the deterministic nature of the current prototype means it cannot evolve beyond its training data. Whether future iterations will incorporate more flexible conversational abilities without compromising reliability remains an open question, and the answer will shape how journalists, CEOs and brands decide to engage with their own virtual selves.