This summer, Alexandru Voica, Synthesia’s head of corporate affairs, sent me a link to their PR team’s newest member—and I was surprised. It was an interactive virtual avatar of him, trained to answer common media questions about Synthesia, such as what the company does and how it works. The day before, I’d been at a forum where a PR professional asked me if I minded using AI-generated text for pitches. But Alexandru’s avatar went further than that—in my view, it was nearly the ultimate expression of AI in public relations.
In September, Synthesia invited me to their new office in New York. Originally headquartered in the UK, Synthesia is a prominent player in the digital human startup space, alongside competitors such as D-ID, HeyGen, and Colossyan. Earlier this year, the company reached a valuation of $4 billion; last year, it reported that its annual recurring revenue had exceeded $100 million.
Synthesia enables enterprises to create interactive training videos using AI avatars, and has recently launched a product called Roleplay Sessions, allowing employees to practice sales scripts and other skills, with AI avatars responding and evaluating their performance.
When I attended the opening event for the new office, they asked if I wanted my own AI avatar, and I agreed without hesitation. Of course, I wanted my own digital twin. That day, I was dressed well and my hair was neat.
Before seeing my digital twin, I never paid much attention to avatars, but I believe they will eventually become a part of everyday online life. I heard that some people create avatars based on their own appearance on Instagram to help produce social content. I found all of this fascinating, which is probably why I now feel comfortable sharing my digital twin with everyone. This is the first time Synthesia has created a digital avatar for a journalist—and the first time it has created one for anyone at all, at least outside of Voica. It was trained on my article titled “Why Venture-Backed Startups Are More Likely to Commit Fraud Than Those Without Venture Funding,” and it will only answer questions about that article.
Next, simply click "Start in a New Window" to begin.
You can ask it questions like:
- Why did I decide to write this article?
- What is this research paper about?
- What did researchers discover?
To create it, I went into a small studio at Synthesia’s office, where they took many photos of me and recorded two minutes of audio. I had to approve the creation of these avatars, and thus, the digital version of Dom was born. They created a personal avatar for me—capable of reading aloud any script I provide—in both glasses and no-glasses versions; they also created two interactive avatars for me—able to converse with me and listen to my voice—again, in both glasses and no-glasses versions.
We selected an article to train this interactive avatar, and a team within the company built this avatar for me, powered by a combination of speech-to-text, video, language, and text-to-speech models. My avatar’s technology stack includes Synthesia’s proprietary video and voice models, but the company also allows customers to choose alternative solutions from other labs, such as Cartesia, ElevenLabs, Google, or OpenAI. Enterprises can also choose to deploy the avatar on any cloud platform they prefer, or pay Synthesia to host it for them.
The speech-to-text model converts spoken words into text, the intelligent agent language model understands the text and takes action accordingly, the text-to-speech model transforms the response into audio, and finally, Synthesia’s proprietary video model animates the avatar to move its mouth as if speaking.
Overall, Synthesia primarily offers three types of products—a video creation and distribution platform with traditional avatars that speak user-input scripts; an agent platform called Sessions, where users can interact with avatars through surveys or role-playing; and an API platform that enables users to integrate Synthesia’s video and voice models with other tech services to build interactive avatars or other types of products.
The Synthesia team spent several days creating my avatar. I first experimented with a personal avatar, entering a fairly ordinary script to hear how the AI voice sounded. I had it talk about how autumn had arrived in New York—my favorite season of the year. The voice was quite accurate, and I was glad it didn’t carry the raspiness from my recording samples.
I showed it to some friends outside the tech world, and they found it both interesting and a bit scary.
Then I showed them my interactive avatar. It is deterministic, meaning it only responds with what it was trained to say—in this case, my venture capital fraud report. I asked it questions, such as where I worked before joining TechCrunch and which part of New York I live in, but each time it brought the conversation back to that article.
My friends thought the voice didn’t sound like me and the avatar wasn’t as realistic as my profile picture, but it was still unsettling enough. My mom called it “amazing”—which is high praise from her. She and my dad kept trying to ask it questions about me that only they would know—but each time, the model refused to answer and redirected them back to the venture capital fraud article.
“I don’t remember giving birth to two of you,” she joked after testing the model.
This experience made me start thinking about the future of journalism. Will people be willing to open the news and see an avatar delivering it? One investor immediately told me no. Today, low-quality AI-generated content has already infiltrated social media and other news-sharing platforms, and the backlash is significant. But others I spoke with weren’t so certain. Could avatars enhance or even replace journalists? Would corporate executives prefer to speak with a journalist’s AI avatar rather than a real person?
What I love about this job is building connections, writing stories, and researching new topics. But the most important part of journalism is trust—and that can never be outsourced to AI.
Outside of journalism, I’m sure the idea of “cloning yourself” would be appealing—never again would you need to catch up on work after holidays or vacations, since there would always be a version of you available to answer questions.
We still need to see how avatar usage will evolve in the U.S. business world. But now I’ve got one myself, and my feelings are mixed. After the initial novelty wore off, I stared at my avatar, waiting after it finished speaking—waiting for what, I’m not sure. Maybe I was waiting for it to blink, say something new, laugh, or simply let me know it understood something.
Because my digital twin is deterministic, it would never do that. But I can imagine how easily people might fall into a bit of AI delusion if a twin were non-deterministic and powered by a chatbot free to improvise and chatter.
I told an investor that no matter what the future holds for digital avatars, I predict our generation—Gen Z—won’t really get used to them. They feel incredibly sci-fi: everything we’ve seen in movies has now become real. But I’d say, compared to humanoid robots, I find digital avatars less unsettling. At least with an avatar, if things get weird, I can always log off.
Before that, check out my profile picture—it’ll give you an overview of all this week’s top stories on our site!