Image by Synthesia

Two articles this week highlighted the complex debate around the companies creating deepfakes, or synthetic media, in which a person’s voice, image or video is replaced by an AI-generated version. Both involved Synthesia, a London-based startup that says you can create “professional AI videos in 15 minutes” by typing in text in over 120 languages.

Synthesia reportedly in talks to raise over $50 million

First of all, yesterday a report from Business Insider said that Synthesia is in talks to raise funds in a deal that could value it at around $1 billion. One source claimed the company could raise between $50 million and $75 million — which is big money in a category that many are pushing back on.

An NPR articleyesterday, for example, said that policymakers can’t keep up with AI-generated deepfakes, which also made headlines this week when the Republican National Committee used AI to create a 30-second ad imagining what President Joe Biden’s second term might look like.

Deepfake tools could ‘fool my family and trick my bank’

Meanwhile, in the Wall Street Journal, columnist Joanna Stern wrote about testing Synthesia to see if the AI tool could make her voice and video work more productive and less like drudgery. Her “AI Joanna,” she said, had its good points: “She never loses her voice, she has outstanding posture and not even a convertible driving 120 mph through a tornado could mess up her hair.”

Unfortunately, she also found that AI Joanna could “fool my family and trick my bank.” She cloned her voice using a different tool by ElevenLabs. While her cloned voice didn’t ultimately deceive her family, it did get around her Chase credit card’s voice biometric system.

Still, Stern said she would continue using these tools. While the video clone and voice weren’t perfect (“lacking the things that make me me“) they can be helpful to save time “to be a real human.”

The authorized, business side of deepfakes

And that’s the point of the other side of the deepfake debate, which maintains that the word “deepfake” implies the unauthorized use of synthetic media and generative AI. But tools like Synthesia, as well as offerings from companies like Hour One, are about the authorized use of this technology for use cases such as business video production.

And that is what investors are interested in: One of Forrester’s top 2023 AI predictions was that 10% of Fortune 500 enterprises will generate content with AI tools. The report mentioned startups such as Hour One and Synthesia which “are using AI to accelerate video content generation.” 

In an interview with VentureBeat last November, Victor Riparbelli, CEO of Synthesia, said that the business side is a “hugely under-appreciated” part of the deepfake debate.

“It’s very interesting how the lens has been very narrow on all the bad things you could do with this technology,” Riparbelli said. “I think what we’ve seen is just more and more interest in this and more and more use cases.”

Riparbelli added that its AI is trained on real actors, and besides users creating videos of themselves, it offers actors’ images and voices as virtual characters clients can choose from to create training, learning, compliance and marketing videos. The actors are paid per video that’s generated with their image and voice.

For enterprise clients, a video created this way becomes a “living video” that they can always go back to and edit, he explained.

And in terms of creating authorized AI versions of users, Riparbelli said this kind of authorization is common in all kinds of licensing agreements that already exist.

“Kim Kardashian has literally licensed her likeness to app developers to build a game that grossed billions of dollars,” said Riparbelli. “Every actor or celebrity licenses their likeness.”