Cover image | Produced by Pencil News

At the start of 2024, Chen Mian's company account was down to its last 4,000 yuan.

After leaving ByteDance to establish the AI image platform LiblibAI, he burned through over 3 million US dollars in less than a year, nearly depleting his funds.

Two and a half years later, the company is gearing up for an IPO.

Recent reports from relevant media indicate that LiblibAI's parent company, YanYu Technology (EVOKEN), is in preliminary talks with advisors regarding a Hong Kong listing. Simultaneously, it is nearing the completion of a new funding round, valuing the company at approximately 3 billion US dollars (about 20.2 billion yuan). Currently, discussions about the listing are still in their early stages, and no final decision has been made.

As of May this year, the company's disclosed Annual Recurring Revenue (ARR) has surpassed 300 million US dollars.

Interestingly, YanYu Technology doesn't train foundational large models; instead, it aggregates multiple models to cater to image and video creation needs. Some may wonder: why is an intermediary company worth 20 billion yuan?

Lu Bin (a pseudonym), an entrepreneur in the AI video sector, told Pencil News that mere video generation tools cannot create long-term value. "What users truly need is a polished result," he said.

- 01 - Why is it worth 20 billion yuan?

YanYu Technology's founder, Chen Mian, born in 1992, graduated from Southeast University. He has worked at Mobike, Didi, MissFresh, and ByteDance.

At ByteDance, he served as the global commercialization leader for CapCut and Jianying, becoming ByteDance's youngest product executive at the 4-1 level (senior management) at the age of 28.

In 2023, he founded LiblibAI, initially as a simple model-sharing community for AI art enthusiasts.

At that time, the full open-sourcing of Stable Diffusion sparked a wave of creativity in image models. However, ordinary users had to find, download, install, and adjust parameters for models to generate high-quality AI images. LiblibAI's role was to bring together models, creators, and generation tools.

It quickly grew into one of China's largest AI material websites and creator communities. According to public data, LiblibAI has amassed over 30 million users. It is said that one in every three designers in China uses LiblibAI.

But being just a "model supermarket" does not guarantee a solid business foundation.

Because models are rapidly becoming commoditized. Today's best model may be replaced by a stronger, cheaper one in a few months.

So YanYu Technology continued to evolve.

In 2025, the company separated professional design needs and launched the design agent Xingliu and Lovart for overseas markets. In March of this year, it launched LibTV, entering the AI video production arena.

These products seem to ride the waves of AI art, design agents, and AI video, but they share a common logic:

Shifting from providing users with models to providing them with completed works.

Today, when users open LibTV, they no longer face a single video model but a workspace. Different models can be integrated into the same creative workflow, allowing creators to avoid redoing their entire production process when switching models.

The three products form a matrix: LiblibAI gathers traffic and creators, Xingliu/Lovart validates design monetization, and LibTV captures the dividends of AI short dramas.

Lu Bin clearly explains this opportunity.

He believes that in the future, there will be more and more models, each with its strengths. The market will thus need a platform or workstation to centralize different models, calling and orchestrating them according to various generation, imaging, and design tasks while preserving users' long-term context, memory, and assets. These are the truly irreplaceable elements.

YanYu Technology's evolution represents a typical path in this round of AI application entrepreneurship: first acquiring users with a high-frequency tool, then gradually transforming the single-point tool into a workbench.

In the capital market, this model has gained recognition. In June of this year, it completed a nearly 300 million US dollars Series B+ funding round, once again setting a new record for domestic AI application-layer single-round financing.

- 02 - Who is paying for AI creation?

The closer models and products are to the production end, the more valuable they become.

Lu Bin told Pencil News that professional creators' core needs can be broken down into three layers:

The first layer is integrated workflows. Previously, producing an AI short drama required switching between Runway for video generation, Midjourney for imaging, and Jianying for editing. Now, a single platform handles the entire process, improving efficiency by at least three times.

The second layer is multi-model collaboration. The same canvas simultaneously calls image, video, music, and text models from different vendors without requiring separate subscriptions or repeated switching.

The third layer is asset precipitation. Scripts, parameters, and workflows from each creation can be saved as templates for direct reuse next time.

"Using industrial production as an analogy," he said, "a single AI tool is like an independent machine tool, while a platform integrates multiple machine tools, conveyor belts, and quality inspection stations into a production line."

Companies that can make money, including YanYu Technology, operate on the third layer. LibTV achieved over 1 million US dollars in daily revenue in its first month, serving over a thousand short drama teams and film and television institutions. By May, its revenue had grown 13 times compared to the first month.

As of May 2026, YanYu Technology's ARR (Annual Recurring Revenue) has exceeded 300 million US dollars (about 2.022 billion yuan). The company's latest performance guidance to investors is that ARR could reach up to 700 million US dollars by the end of the year (about 4.72 billion yuan).

What does 700 million US dollars mean? Goldman Sachs predicts that Kuaishou's Kling AI will reach 1 billion US dollars in ARR by the end of the year. An application company that does not build models is approaching the revenue scale of model-building vendors.

Just yesterday, on August 17, the U.S. AI content platform Higgsfield announced the completion of 400 million US dollars in funding, reaching a valuation of 5.4 billion US dollars, four times its valuation about half a year ago. Its annualized revenue has reached 700 million US dollars, with over 30 million users.

More critically, its customer structure has changed. In January of this year, enterprise customers contributed less than a quarter of Higgsfield's revenue; now, most of its revenue comes from enterprises. Its customers are shifting from individual users trying out the technology to companies that need to produce advertising and marketing videos in bulk daily.

Similarly, Kuaishou's Kling AI achieved nearly 500 million US dollars in ARR in less than a year, with 70% coming from overseas markets. Its paying users are professional creators such as self-media creators and advertising and marketing practitioners. In the TV drama "Peaceful Years," Kling participated in creating some virtual scenes and visual effects shots. In the overseas series "House of David," it generated hundreds of shots, including complex war scenes.

This is precisely YanYu Technology's core strategy—serving professional creators. Among LiblibAI's 30 million users, core paying users are designers and creative professionals. Xingliu targets brand and commercial design audiences. LibTV directly serves short drama teams, film and television production institutions, and advertising companies—all content-dependent professionals.

- 03 - What to do when big firms enter?

However, in Lu Bin's view, merely occupying the creative production line is not the endpoint.

He has an even more aggressive prediction: AI video may ultimately form an independent content format.

He calls it AIDEO: a combination of AI and video. Just as smartphones lowered the barriers to shooting and dissemination, spawning short videos, generative AI may further lower the production barriers for professional imagery.

Producing a video of decent quality previously required cameras, locations, actors, lighting, and editing. In the future, models may handle a significant portion of these tasks.

This will lead to a result: a sharp increase in content supply.

When content is no longer scarce, the real opportunity lies not just in helping creators produce but also in assisting with promotion, distribution, and ultimately forming a commercial closed loop.

Currently, YanYu Technology's survival logic is essentially a "time window strategy": during the period when big firms have not fully penetrated niche scenarios and model capabilities have not yet trickled down to segmentation (fine-grained) applications, it uses its first-mover advantage to acquire users, generate revenue, and scale up.

But the window is narrowing.

Take the design agent sector as an example. In July 2026, Tencent fully launched its first self-developed creative agent, Miora, directly targeting brand design and film and television creativity—the core sectors of Xingliu/Lovart.

The AI video sector is even more competitive. ByteDance's Seedance/Jimeng, Kuaishou's Kling, and Alibaba's Wanjing Yike/HappyHorse dominate over 70% of the market revenue at the AI video model layer. These big firms have models, traffic, cloud services, and computing power—once they decide to enter the application layer, how much time will startups have?

Then there's the issue of gross margins. Large model companies can achieve gross margins of over 60%, but AI application products generally have gross margins below 30%. Chen Mian himself admits that this figure is "currently normal," but will the capital markets continue to buy in?

Lu Bin believes that the real barrier lies not in the tool layer but in scenario depth and creator ecosystems—whether you can enable professional creators to settle their work habits, accumulate material assets, and complete the full link (chain) from creation to monetization on your platform.

YanYu Technology's continuous launch of new products is essentially answering the same question: When big firms enter the field, what will keep you relevant?

The current answer is—speed. Over three years, it has launched multiple products, each rushing in as soon as a new window opens, quickly acquiring users, achieving commercialization, scaling up, and then finding the next window before big firms catch up.

Can this strategy be sustained? Lu Bin is optimistic: "A general-purpose platform may not emerge. Instead, highly personalized platforms could appear. Everyone's work scenarios and video creation scenarios are different."

If this prediction holds, then the AI application layer will not see a single dominant player. Seven, eight, or even ten companies could grow in different scenarios. What YanYu Technology needs to do is move fast enough and dig deep enough in its chosen track (track).

The value coordinates of the AI application industry are quietly shifting: from generating content to managing creation; from selling tools to delivering results; and from delivering results to mastering the creator's business.

The content of this article is for reference only and does not constitute any investment advice. This article also references reports from 36Kr, TMTPost, and others, with thanks.