Yingmu Technology: How a Group of Gen Zers Is Using 3D Large Models to Tap into a 10-Billion-Yuan Market

On June 23, 2026, Hyper3D, a 3D generative AI company, announced the completion of a new round of funding worth several hundred million yuanfinancing, with co-lead investors including Cathay Capital and Shanghai SDIC Pioneer, and continued participation from existing shareholders such as Sequoia China, Meituan Dragon Ball, ByteDance, and BlueRun Ventures; LightSource Capital served as the exclusive financial advisor.

This marks the third round of funding that Yingmu Technology has secured since 2025. In less than a year and a half, this company—founded just six years ago—has raised more than 100 million yuan in cumulative funding, with investors including top global venture capital firms, leading Chinese internet strategic investors, and state-level industrial funds.

“It’s rare in the current market environment for a 3D-generation company to secure three funding rounds within a year, with its valuation rising in each round,” an investor close to the deal told a reporter. ”There’s one key reason for this: they’ve chosen the right technological path and have already established a viable business model.”

It All Started with a ”Setback”: Four Young People Who Emerged from a Lab at ShanghaiTech University

The story goes back to June 2020.

At that time, in the MARS laboratory at ShanghaiTech University, 23-year-old Wu Di, 21-year-old Zhang Qixuan, and four other fellow students, armed with a self-developed ”Dome Light Field” micrometer-level facial scanning system—the only one of its kind in Asia and one of only two in the world—registered and established Yingmou Technology in the Shanghai Free Trade Zone.

During the first two years of the startup, the team seemed to be on a smooth path: they used hardware to scan and capture high-precision 3D facial data, which they then processed with software to generate hyper-realisticdigital person. In November 2022, Yingmou launched China’s first in-house developed hyper-realistic digital human system, which at one point secured orders from AAA game companies and film and television projects.

But soon, the team encountered its first major setback.

“One of the first products we developed was a pure lighting tool that allowed users to adjust the light source by uploading photos. Technically, it worked fine, but it was completely incompatible with Unity—the system widely used in the film, television, and gaming industries—so virtually no one bought it.” Recalling this experience in an interview, Wu Di said, ”The most common mistake entrepreneurs with a research background make is to overhype their experimental results while overlooking what customers truly need.”

This ”setback” directly led the team to make a key decision:‌Cut all 2D business lines and go all-in on native 3D generation.‌

The Debate Over Technical Approaches: Why Yingmou Chose the Most Difficult Path

Around 2023, the mainstream approach in the field of AI-generated 3D was ”2D-to-3D”—where large language models infer 3D structures from 2D images. This approach has a short development cycle and requires little data, but the generated results are inaccurate and unpredictable; the industry jokingly refers to it as ”drawing cards at random.”

Ying Mou chose a completely different path:‌Native 3D Large Language Models‌—Enable AI to directly understand and generate 3D structures, bypassing 2D intermediate formats.

“If a model has already been pre-trained, it becomes very difficult to add controllability later on. That’s why we designed controllability into the model during the pre-training phase,” explained CTO Zhang Qixuan.

In 2024, the team officially released the world’s first native 3D large-scale model framework‌CLAY‌, and on that basis, launch a production-ready model‌Hyper3D Rodin‌. The related paper was nominated for the SIGGRAPH Best Paper Award, and Yingmou became the first team from mainland China to be selected for the conference’s Real-time Live! segment in its 50-year history.

In June 2025, the team released‌Rodin Gen-2.5‌, introducing the ”Chain of Thought” logic of large language models into the 3D field for the first time, achieving generation capabilities on the scale of tens of millions of faces, and outputting models with millions of faces in as little as 4 seconds.

Compared with mainstream solutions in the industry, Yingmou’s technological advantages are reflected in several key metrics:

• ‌Training costs‌: Only about one-fifth of the industry average;
• ‌Generative Control‌: Supports natural language editing, recursive segmentation, and 3D ControlNet, rather than ”opening a mystery box”;
• ‌Industrial Availability‌: The geometry is scientifically sound, the routing is well-organized, and the material accuracy meets production-grade standards—not just for demo purposes.

“We’re not really interested in cramming everything into a ”world model,”” Wu Di said with rare restraint when discussing industry buzzwords. “Our focus is very clear—to develop production-ready 3D generation tools.”

Commercial Performance Report: Overseas Revenue Accounts for 80% of Total Revenue; B2B Clients Include Global Giants

Unlike most AI companies, which focus on consumer traffic, Yingmou has centered its commercialization efforts on the enterprise market from the very beginning.

As of now, the company‌Overseas revenue accounts for approximately 80%‌, with a truly impressive client list:

In terms of data, during its first month after launch, Rodin Gen-2.5 saw month-over-month growth in both the number of subscribers and ARR (annual recurring revenue) that exceeded‌400%‌.

“Our overall positioning is somewhat similar to Anthropic’s,” said Wu Di. ”Our B2B business is our most important growth engine. Currently, the number of B2B clients and the revenue we’ve accumulated exceed the combined total of all other companies in the industry.”

Team Profile: A 60-member ”Gen Z” team; 70%’s research achievements are ready for commercialization

As of June 2026, the Yingmou team consisted of approximately‌60 people‌, two-thirds of whom are R&D personnel, with most of the core members born in the 2000s.

In terms of management, Wu Di compares it to DeepSeek’s organizational logic: ”Anyone with an idea can propose it; we discuss it, and if we think it’s feasible, we go ahead and do it. With the help of AI, we can complete a preliminary validation experiment in just one week.”

The team operates with a highly modular division of labor—with dedicated specialists for specific areas such as UV mapping, materials, and topology—and has established a streamlined, rapid validation process. According to publicly available information, within the algorithm team,‌One in every two people has received or been nominated for a Best Paper Award, and approximately 70% of research output can be directly translated into product features.‌.

“In the age of AI, algorithms are the domain of young people,” said Wu Di. ”We’re a very flat team; we don’t rely on experience, because this is a brand-new field—if we did, we’d probably still be working on 2D upscaling.”

The Logic of Capital: Why Three Rounds of Funding in One Year?

A Look Back at Yingmu's Funding Timeline:

Caihui Fund, which has joined this round of financing, is one of Europe’s largest technology investment funds, while Shanghai SDIC Pioneer represents Shanghai’s state-owned industrial capital. The simultaneous entry of both industrial capital and state strategic capital is viewed by the industry as a strong endorsement of Yingmou’s technological approach and commercialization capabilities.

“The 3D generation sector is consolidating,” Wu Di observed. ”Consolidation means that, in the end, only a handful of teams that have truly solved industry problems will remain in the game.”

Industry Insight: 3D Generation Enters a Survival-of-the-Fittest Phase—Where Does Yingmou’s Competitive Advantage Lie?

Since 2025, the 3D generation sector has seen significant acceleration. ByteDance launched Seed3D 2.0, Tencent’s Hunyuan released HY-World 2.0, VAST secured nearly $200 million in funding, and NVIDIA and OpenAI have also entered the market.

However, several industry analysts told this reporter that the sector is shifting from a period of ”diverse growth” to one of ”survival of the fittest,” with the key distinction lying in two points:‌The Correctness of the Technical Approach‌and‌The Authenticity of the Business Ecosystem‌.

“Many teams are still working on demos that convert 2D to 3D, and they have a long way to go before achieving true industrial-grade usability. ” said a 3D industry investor who wished to remain anonymous. ”Yingmou’s competitive advantage doesn’t lie in any single technology, but in the fact that they chose the native 3D path from day one and have already proven themselves with clients like NVIDIA and Lowe’s.”

According to reports, Yingmou’s next strategic focus will be on three areas: first, continuously improving the generation quality of its underlying models; second, expanding editing capabilities and agent-based workflows; and third, deepening its implementation in vertical scenarios such as gaming, e-commerce, 3D printing, and embodied intelligence.

In December 2025, Yingmou signed an agreement with the Taihu Town Government in Tongzhou, Beijing, to focus on AI travel services, 3D generation technology, and AI-native data storage—a move seen as a sign that the company is expanding into the broader AI infrastructure layer.

‌From a failed startup in a lab at ShanghaiTech University to becoming the technical star on NVIDIA’s CES stage, Yingmou Technology has taken six years to traverse the most challenging—yet also the most solid—path. In the field of 3D generation, the fastest runner may not necessarily be the one who survives to the end—but the one who runs the right course might just make it.