What is Liblib's moat?

Article author and source: 36Kr

Recently, the AI video sector has secured another major funding round. Evoken, the parent company of Liblib, completed a B+ round of nearly $300 million, led jointly by Granite Asia, Tencent, and Shunwei Capital. The company’s post-money valuation exceeds $2 billion, making it a new unicorn in the AI applications space.

Image source: Weibo screenshot

This marks another company in the AI video generation space reaching a $300 million funding milestone, following Aishi Technology. However, unlike Aishi Technology, Yanyu Technology is a product company that has grown rapidly by aggregating mainstream models available on the market for designers and creators, without developing its own models.

As large model iterations accelerate, a single model update may overwrite most of a product’s features—whether AI application layers still have genuine moats has been a hot topic in the industry. Against this backdrop, why has Yanyu Technology garnered capital’s favor despite the trend? And does its product truly possess the competitiveness to withstand model iterations?

01. Byte Halo and the Tailwind of Short Dramas

When discussing a company’s ability to attract capital, its founder is always central. Chen Mian, founder of Yanyu Technology, has worked at several major tech companies, progressing from Tencent, 360, Baidu, and Didi to ByteDance. Before launching his startup, Chen served as the Global Head of Monetization for CapCut and Jianying at ByteDance. Born in 1992, he became the youngest Level 4-1 employee in ByteDance at the time.

ByteDance’s background may have made Chen Mian more attractive to investors. According to Tianyancha, the company secured its angel round of funding from Source Code Capital, Gao榕 Capital, and Jinsha Ventures just two months after its establishment. As of June this year, Yanyu Technology has completed six rounds of funding, with Tencent, Ant Group, Sequoia Capital, and Shunwei Capital joining in its recent B+ round.

Image source: Tianyancha screenshot

Beyond Chen Mian’s background, the rapid growth of Yanyu Technology in terms of spending has been supported by its user base and revenue trends. Yanyu Technology’s core products include Liblib AI, an AI creation community launched in 2023; Xingliu and Lovart (overseas version), AI design agents launched in 2025; and LibTV, an AI video creation platform launched in March this year. Among these, Liblib AI has accumulated over 30 million users, making it one of China’s largest AI asset websites and creator communities.

Based on the disclosed ARR, the company’s total annualized revenue has surpassed $300 million. After five months of launch, Lovart achieved an ARR of over $80 million; while no specific figures are provided for LibTV, Yanyu Technology stated that LibTV’s monthly revenue increased more than 13-fold within two months of its launch.

Lovart has gained strong popularity overseas by addressing the pain point of fragmented design tools. Traditional AI design tools only generate individual images—after users input prompts and receive an image, they must manually switch to Photoshop for editing and Figma for layout. With Lovart, users can generate complete design solutions in one go.

Liblib is the same—it aggregates multiple models, providing video creators with a more streamlined workflow, and this convenience is the foundation upon which tool-type agents thrive.

Image source: Screenshot from LibLibAI's official website

"Qu Jie Shang Ye" has noticed that Yanyu Technology rapidly iterates its products. When LibTV first launched, some users new to AI editing said they couldn't learn fast enough to keep up with the software's updates.

Unlike Lovart, LibTV was fortunate to launch at the perfect time, coinciding with the industry shift from live-action to animated short films. After this year’s Chinese New Year, Hongguo Short Films aggressively entered the AI short film space, prompting numerous short film producers to seek reliable video generation tools. To date, LibTV has already acquired a substantial user base among short film creators—who currently represent the core consumer group in the video generation field. In addition, teams requiring high-frequency production of visual content and marketing materials, such as A-list advertising agencies and brand creative departments, are also key users of LibTV.

However, compared to a streamlined workflow, LibTV's greatest competitive advantage is its price.

Due to scarce computing power and a surge in users, Mind4 adjusted its pricing three times in April, causing the cost of video generation to nearly sextuple. As a result, LibTV, which also offers access to "Seedance 2.0," has become a popular cost-effective alternative for many AI video professionals.

According to the LibTV official website, LibTV's creative members are currently divided into different tiers—Standard, Advanced, and Premium—each with varying video generation limits, priced between 569 and 8,499 yuan per year. Some editors have noted that since Ji Meng raised its prices in April, their team has gradually switched to LibTV for video production. Although there are fewer editing tutorials and it’s harder to find answers to problems, the lower cost and lack of waiting times are the most important factors.

Source: Screenshot from LibTV

Some comic creators have noted that LibTV could generate one minute of content for as little as 20 yuan; it’s very difficult to get a lower price on Ji Meng, and such a low cost might only be possible under strict conditions like off-peak computing power or limited积分.

02. It’s Hard to Be a Middleman

As a "model migrator," why can Liblib offer lower prices than the original platform?

This touches on the commercial essence of aggregation platforms. Liblib provides creators with access through API interfaces from model providers; all computational power consumed on Liblib originates from the model providers’ computing centers. In other words, users’ spending on Liblib primarily depends on the cost of purchasing computational power, with Liblib earning the margin as an intermediary.

If Liblib commits to consuming a certain volume of computing power in Token terms annually, it may receive a lower discount rate; however, the discount won't be so substantial that LibTV could sell to users at a price far below Ji Meng and still make a profit.

To maintain "high cost-effectiveness" on the C-side, there are only two options: either subsidize users out of pocket or seek cheaper channels, such as intermediary APIs, to obtain computing power quotas.

This may also be why Yanyu Technology has raised funds frequently. Without the capability to develop models, it primarily relies on subsidies to retain users, much like the cash-burning battles in the early days of mobile internet, where platforms used subsidies to gain scale and cultivate user habits.

However, if creators choose Liblib due to an increase in Dream's pricing, they may also abandon Liblib if a cheaper platform emerges or if Dream lowers its prices.

It is also worth noting that platforms often overlook content compliance during the phase of capturing market share. In April of this year, China Central Television Finance reported that LiblibAI was experiencing an “AI-generated explicit content” issue, where users could bypass moderation by using subtle prompts to generate prohibited content. Subsequently, LiblibAI apologized and stated that technical fixes had been completed.

For AI tools, it is essential to strengthen prevention of security risks and compliance issues.

Image source: Weibo screenshot

A dedicated investor in hard tech bluntly stated that the moat of AI aggregation platforms is shallow; once major companies open their APIs, lower prices, or launch superior native applications, these platforms can be easily replaced. More critically, against the current intensity and speed of model iterations, the differentiated advantages of these aggregation platforms are highly fragile—features refined over six months can be rendered obsolete by a single model update.

This is not just Liblib’s issue—most AI application-layer tools face similar structural challenges, as their product value is heavily dependent on the capabilities of upstream models.

Whether AI will kill software has also been a repeatedly discussed topic this year.

In January this year, Anthropic launched Claude’s tool Cowork and industry-specific plugins covering areas such as legal, financial, and sales services, replacing some vertical SaaS software and triggering multiple declines in Nasdaq software stocks. Anthropic’s CEO, Dario Amodei, has previously expressed a similar view: once large models mature, software capabilities—such as video generation and graphic design—will be embedded within general-purpose foundational models, eliminating the need to develop and sell third-party tools independently.

However, NVIDIA CEO Jensen Huang has publicly opposed the view that “AI will drain software,” arguing that agents will be built on top of enterprise systems and structured data. Software providers will not be replaced but rather need to transform.

The relationship between software and AI has not yet reached a consensus within the industry; as a result, some have characterized platforms like Liblib as "intermediate products during the period before models converge." LibTV has achieved more than a 13-fold increase in monthly revenue within its window of opportunity, making it a rare winner—being "fast" itself is a scarce capability.

However, whether Liblib’s growth can be sustained largely depends on continued funding and its ability to evolve from a “model replicator” into a must-have creative infrastructure for users before the window of opportunity closes.