Over the past six years, the company has built a dizzying library of avatars. They’re available in different genders, skin tones, and uniforms. There are hipsters and call center workers. Santa is available in multiple ethnicities. Within Synthesia’s platform, clients can customize the language their avatars speak, their accents, even at what point in a script they raise their eyebrows. Riparbelli says his favorite is Alex, a classically pretty but unremarkable avatar who looks to be in her mid-twenties and has mid-length brown hair. There is a real human version of Alex who’s out there wandering the streets somewhere. Synthesia trains its algorithms on footage of actors filmed in its own production studios.
Owning that data is a big draw to investors. “Basically what all their algorithms need is 3D data, because it’s all about understanding how humans are moving, how they are talking,” says Philippe Botteri, partner at venture capital firm Accel, which led Synthesia’s latest funding round. “And for that, you need a very specific set of data that is not available.”
Today, Riparbelli is the rare type of founder who can talk about his vision for game-changing technology while also doing the grunt work of signing up current-day clients. “Utility over novelty” is Synthesia’s internal company mantra, he explains. “It’s very important for us to build technology for real markets that have actual business value, not just to produce cool tech demos.” Right now, the company claims it has 50,000 customers. But Riparbelli also wants to develop technology that can enable anyone to use text to describe a video scene and watch AI generate it. “Imagine you had a movie set with people in front of you, and you got to tell them what to do,” Riparbelli says. “That’s how I imagine the technology is going to work.”
But Synthesia’s technology has a way to go first. Right now, the R&D team is focused on what Ripbarbelli calls the “fundamental AI tech.” The company’s avatars are trapped in invisible straightjackets, unable to move their arms. And unsurprisingly, letting fake humans loose into the wild has not been without its problems. For several years, Synthesia’s avatars—especially an authoritative-looking deepfake the company calls Jason—have been impersonating news anchors on social media, reading scripts that have been written to spread disinformation.
In December 2021, Jason appeared on a Facebook page associated with politics in Mali, making allegations that fact-checkers called false about France’s involvement in local politics. Then in late 2022, there he was again, condemning US failure to act against gun violence—with the social media analysis firm Graphika linking the video back to a pro-China bot network. In January this year, people noticed Synthesia avatars expressing support for a military coup in Burkina Faso. And by March, fact-checkers were raising the alarm about another Synthesia-linked video circulating in Venezuela—this time it was the avatar Darren arguing that claims of widespread poverty in the oil-rich country had been exaggerated. The video was boosted by accounts supportive of President Nicolas Maduro. In April, California’s financial regulator found avatar Gary being used in a crypto scam, pretending he was a legitimate CEO.