Synthesia builds an interactive PR avatar, then makes one for a TechCrunch reporter

Synthesia has built an interactive digital replica of its own communications chief that can answer press questions in real time — and, to demonstrate the capability, it made one for a TechCrunch journalist as well.

According to TechCrunch, Alexandru Voica, Synthesia's head of corporate affairs, sent the publication a link this summer to a conversational clone of himself, trained to field common questions about the company. When the reporter later visited Synthesia's newly opened New York office in September, the company offered to build a "digital twin" of them too — which TechCrunch says is the first avatar the company has made for anyone outside Voica himself.

How the interactive avatars work

The interactive avatar is assembled from four model types. A speech-to-text model transcribes what a person says, an agentic language model interprets the text and decides what actions to take, a text-to-speech model converts the response into audio, and a video model built by Synthesia animates the avatar as it talks. TechCrunch reports that customers can swap in components from other labs — Cartesia, ElevenLabs, Google or OpenAI — and can either host the avatars on whatever cloud they choose or pay Synthesia to host them.

Creating one is fast. The journalist sat in a small studio inside Synthesia's office, posed for a series of photos, recorded two minutes of audio and gave explicit consent. Within a couple of days the team delivered a personal avatar (one that simply reads whatever script it is given) plus interactive versions, each rendered with and without glasses.

Grounded in one story, on purpose

The interactive clone is deterministic: it says only what it was trained to say. Synthesia grounded the journalist's avatar in a single article about why venture-backed startups commit more fraud than companies without VC funding, and the avatar steered every off-topic question — including personal ones from the reporter's parents — back to that piece.

TechCrunch notes that the voice and likeness of the interactive version were less convincing than the script-reading personal avatar, though close enough to unsettle non-technical friends who saw it. The reporter also raised a concern about nondeterministic clones powered by chatbots free to improvise: people could slip into an unhealthy relationship with a replica that keeps generating new things to say.

The business behind the demo

Synthesia, originally based in the UK, sells AI avatars for enterprise training video and competes with firms such as D-ID, HeyGen and Colossyan. Per TechCrunch, it reached a $4 billion valuation earlier this year and said last year that it had passed $100 million in annual recurring revenue.

Its portfolio spans three product lines. A video creation and distribution platform offers classic avatars that read typed scripts. An agentic platform called Sessions supports surveys and interactive exercises, including a Roleplay Sessions feature that lets employees rehearse sales pitches against a responsive avatar that scores their answers. An API platform lets customers combine Synthesia's video and voice models with other services to build their own interactive avatars or other products.

Why it matters

A communications chief who can take routine reporter questions at any hour, indefinitely, is a concrete example of real-time conversational avatars moving into corporate communications — not as a novelty demo, but as an always-on extension of a PR team.

The demo also shows the guardrail that makes such deployments viable: grounding the avatar in approved material so it cannot wander into unverified claims. That addresses hallucination, but it creates a disclosure problem, since nothing about the interaction forces the avatar to announce that it is not a person.

The capture requirements are minimal — a handful of photos, two minutes of voice, a few days of build time — which lowers the barrier for any executive or public figure wanting a clone. Whether audiences accept them is another matter: as the TechCrunch reporter observed, trust is the core of journalism and seems impossible to outsource to a machine. For answering the same company FAQs again and again, though, that may not be the deciding factor.