When OpenAI previewed Sora in February 2024, the race to turn text into video looked like a contest for labs with billion-dollar budgets. A small Beijing team had already shipped its own generator weeks earlier — and its founder had spent years, inside ByteDance, learning exactly how to make machines understand what a video actually is.
Wang Changhu (王长虎) is the founder and chief executive of AIsphere (爱诗科技), the company behind PixVerse, a consumer text-to-video (文生视频) and image-to-video (图生视频) app with more than 100 million registered users. He is one of the clearest examples of a Chinese AI builder who came not from a research lab or a Western PhD pipeline, but from the trenches of a consumer product used by a billion people a day.
Key takeaways
- Wang Changhu (王长虎), born in 1982, graduated from the University of Science and Technology of China (中国科学技术大学) in 2009.
- He spent 2017–2021 at ByteDance (字节跳动) as Director of Visual Technology, building the computer-vision stack behind Douyin and TikTok.
- He founded AIsphere (爱诗科技) in April 2023; its product PixVerse launched globally in early 2024 and passed 100 million registered users within about two years.
- The company raised a large Series C in 2026, with reporting putting the round near US$300 million (≈ HK$2.3B) and a later extension above a US$2 billion (≈ HK$15.6B) valuation.
The engineer who labeled a billion videos
Wang's career is a study in scale. After university he joined Microsoft Research Asia (微软亚洲研究院) as a principal researcher, publishing on computer vision, video understanding and multimedia retrieval. In 2017 he moved to ByteDance (字节跳动), where he became the company's Director of Visual Technology. His mandate was foundational: build the visual algorithms that let Douyin and TikTok understand, tag and recommend the flood of video their users uploaded.
That work is the through-line of his story. Recommendation systems at TikTok's scale do not need a single perfect model; they need efficient, repeatable ways to label enormous volumes of visual data and feed it back into the product. Wang has said, and his co-founder has repeated publicly, that PixVerse's real edge is not data volume but data labeling — knowing what a frame contains, and why. It is the same instinct that powered a consumer app used by a billion people, and it travels surprisingly well into generation.
What AIsphere actually builds
AIsphere (爱诗科技), incorporated in Beijing in April 2023, organized its product around a simple bet: video generation would move from a research demo to a consumer tool faster than incumbents expected. PixVerse launched globally in January 2024, offering text-to-video (文生视频) and image-to-video (图生视频) generation, then iterated aggressively — a diffusion-transformer architecture in early 2024, followed by rapid releases adding character consistency, camera control and faster rendering.
By late 2025 the company said its user base had crossed 100 million registered accounts, with annual recurring revenue reportedly above US$40 million. Those numbers, if accurate, place PixVerse among the most-used consumer AI video apps worldwide, in the same conversation as Runway, Kling (可灵) and Sora. The product now spans three model families: a V-Series for consumers and developers, a C-Series for professional film and advertising, and an R-Series "world model" the company introduced in early 2026 for real-time, 1080p generation aimed at games and simulation.
A funding story that tracks the category
PixVerse's rise coincided with a wave of capital into AI video. The company closed a Series C in early 2026 that reporting tied to roughly US$300 million (≈ HK$2.3B), led by CDH Investments (鼎晖), at a valuation around US$1 billion — unicorn status. A later extension pushed the total toward US$439 million (≈ HK$3.4B) and a valuation above US$2 billion (≈ HK$15.6B), with Alibaba among the backers. The figures come from press reporting and the company's own announcements; treat them as reported rather than audited.
Wang's own framing is about efficiency, not spend. In interviews he has argued that AIsphere reached competitive quality using only a fraction — "tens of percent, even one percent" — of the resources rival labs consumed, a claim grounded in the GPU-discipline habit he says his team built managing large compute clusters at ByteDance. It is a familiar China-AI narrative, and in PixVerse's case it is at least partly measurable in shipping cadence: the team releases new model versions on a roughly quarterly rhythm, a pace that keeps the product in front of users between the marquee launches.
The competitive field is crowded and consolidating. ByteDance runs its own Seedance model, Kuaishou ships Kling, and OpenAI's Sora set the public benchmark. PixVerse's answer has been to lean on the freemium consumer app for distribution while pushing the R-Series toward use cases — game worlds, simulation — where few rivals ship production-ready tools today.
What readers can do now
- Test the labeling-accuracy thesis yourself. Generate the same prompt on PixVerse and a competitor; the differentiator Wang cites is not raw data but how well the model interprets intent — judge output on instruction-following, not novelty.
- Separate registered users from revenue. PixVerse reports 100M+ registered users and an ARR above US$40M; a freemium app can have huge reach with thin conversion. Watch ARR growth, not just user counts.
- Track the R-Series world model as the real bet. Consumer video is crowded; the company's longer game is real-time generation for games and simulation, where few rivals ship today.
Honest limitations
This profile is built from company announcements, Baidu Baike and founder bios, and English-language reporting (TechCrunch, Bloomberg-sourced coverage, industry press). Core facts — Wang's USTC education, Microsoft Research Asia and ByteDance tenure, the April 2023 founding of AIsphere, and PixVerse's early-2024 launch — are corroborated across multiple independent sources. User and revenue figures (100M+ registered, US$40M+ ARR) are the company's own disclosures, reported by press; I present them as claimed. Funding amounts and valuations are from press reporting and may be revised. PixVerse's internal roadmap, unreleased models, and the technical details of its R-Series architecture are outside what public sources confirm. No security or commercial endorsement is implied.