A marketing manager in Ohio records herself once. Minutes later she appears to deliver the same pitch in Spanish, Japanese, and German, her lips moving in sync with each language. No camera crew, no translators, no reshoots. The company behind that trick, HeyGen, was started by two graduates of Tongji University (同济大学) — and it has become one of the clearest examples of a China-rooted team building a synthetic-video business for the entire world.
The product is a digital human (数字人): a photo-realistic avatar that recites a script in dozens of languages while its mouth and expression match the audio. It sounds like science fiction; it ships as a subscription.
The founders and the pivot
HeyGen was founded in 2020 by Joshua Xu (徐卓) and Wayne Liang (梁望), initially under the name Surreal, then Movio, before settling on HeyGen. Both hold degrees from Tongji University and later Carnegie Mellon; Xu was an early engineering lead at Snapchat, Liang a product designer at ByteDance. The founding team's early backers included Sequoia China (now HongShan) and ZhenFund (真格基金) — a thoroughly China-side starting point for what became a Los Angeles-headquartered company.
The breakout moment came in late 2023, when a video-translation feature — the "same person speaks another language" demo — spread across social feeds. The viral hook was not a model benchmark; it was the eerie usefulness of a single recording becoming many localized videos.
The business, in verified numbers
The funding and scale figures that hold up across Bloomberg, SCMP, Forbes, and Wikipedia are specific:
- In November 2023, HeyGen raised US$5.6 million from Conviction (Sarah Guo's firm) with a team of about 25 people.
- In June 2024, it raised US$60 million led by Benchmark, at a US$500 million valuation, reporting more than 40,000 customers and an annualized recurring revenue of about US$35 million.
- Later in 2024 it was named to Fast Company's "Next Big Things in Tech" and Inc.'s "Best in Business" lists.
These are the numbers I treat as solid. Later, Chinese-language coverage has cited far larger ARR and user counts (some above US$100 million ARR and tens of millions of registered users); those single-source figures are plausible but not independently corroborated here, so I present the June 2024 set as the verified baseline and treat the rest as directional.
Why avatars, not "video AI"
HeyGen's positioning is deliberately narrow. It does not try to be Sora or Runway — open-ended generators of any scene. Instead it preserves a real person's face, voice, and mannerisms and replays them in another language or script. The company describes itself as "identity-first," meaning the human stays the subject rather than being replaced by a synthetic stranger.
That choice has compliance upside. Every custom avatar must be verified by the person depicted through an on-camera identity check, and the platform blocks political campaign content and requires verbal consent plus a spoken password before generating. Those guardrails matter because the same lip-sync technology sits close to deepfake territory; HeyGen's answer is to bind the avatar to a verified human rather than ship faces freely.
The China angle, stated plainly
The "China angle" here is not government or policy — it is pedigree and capital. The founders are Chinese-educated, one cut his teeth at a major Chinese tech company, and the earliest money came from Chinese venture firms. The product, however, is global by design: a U.S. headquarters, clients in marketing and training, and a value proposition aimed at any company that needs to localize video at scale.
One reporting detail worth noting without overdramatizing: coverage of the 2024 round described the company as having moved away from mainland-China investors in its cap table. That is a structural funding fact, not a political stance, and it reflects how cross-border AI startups navigate where their money sits.
Where it fits in the landscape
HeyGen competes directly with digital-human platforms like Synthesia and D-ID, and complements rather than rivals general video models. Its strength is the boring-but-valuable enterprise workflow: training videos, product explainers, and localized ads where a consistent on-screen presenter matters more than cinematic novelty.
What readers can do now
- Test the localization use case, not the novelty. If you produce any training or marketing video that needs to reach more than one language, the translate-and-lip-sync workflow is the feature worth piloting — measure time saved against a traditional dubbing vendor.
- Demand identity verification. When evaluating any avatar tool, check whether it requires the depicted person to consent and verify. The presence of that gate is a decent proxy for how seriously a vendor treats misuse.
- Keep a human in the loop for claims. Avatars make polished video cheap; that lowers the cost of saying something wrong. Review script accuracy and localization nuance before a generated spokesperson "speaks" to customers.
Honest limitations
This article relies on funding and scale figures reported by Bloomberg, SCMP, Forbes, and Wikipedia — the June 2024 round (US$60M, US$500M valuation, 40k customers, ~US$35M ARR) is the verified core. Later, larger ARR and user numbers from Chinese-language outlets are single-sourced here and are therefore presented as directional rather than confirmed.
I have not used HeyGen's product hands-on, so the description of verification flows and guardrails comes from the company's stated policies and secondary reporting, not personal testing. Two dimensions are uncovered: the underlying speech and lip-sync models are proprietary and their training data is not public, so bias or voice-cloning-accuracy questions cannot be assessed from available sources; and the long-term regulatory treatment of avatar/lip-sync tools in different markets remains unsettled, which affects how durable the current business model is.