Recently, the field of AI video generation has witnessed a series of significant developments.
First, Beijing Aish Technology Co., Ltd. officially announced the completion of a $60 million (approximately 430 million RMB) Series B financing round, breaking the record for the single – largest financing amount in the domestic video generation field.
Then, another star startup, Shengshu Technology, also announced the completion of a Series A financing round worth hundreds of millions of RMB, with participation from well – known investment institutions such as Baidu and Qiming Venture Partners.
As for the established player Kuaishou Keling, it has launched a new digital human function.
It is understood that it is fundamentally different from the common AI anchors in the past. Now, users only need a character image and a piece of text or audio to generate a 1080p/48FPS digital human video of up to 1 minute. At the same time, it can achieve precise lip – sync and fine control of emotional movements, making the character “both physically and spiritually vivid”…
So far, from the replenishment of capital “ammunition” to technological innovation breakthroughs, the AI video generation industry has prepared for the next explosion.
Then, in the future, who will be able to lead the way?
The “Divergence” in AI Video Generation: Small Companies Focus on Products, Large Companies Focus on Ecosystems
From February last year, when OpenAI released the artificial intelligence text – to – video large – model Sora, shocking the industry, to now, the AI video generation track has gone through nearly two years of explosive growth.
Although the overall pattern is not yet fully clear, in terms of model performance, the players in the first echelon have basically stabilized.
Combining two widely recognized industry lists (the international evaluation institution Artificial Analysis and the AGI – Eval jointly released by several top domestic universities and research institutions), among the top 10 global video generation models, except for Google, the rest are all domestic players.
There are mainly two camps: One is the domestic Internet giants, including Kuaishou Keling, ByteDance, and Alibaba.
The other is the domestic startups, mainly Beijing’s Aish Technology (PixVerse V5) and Shanghai’s MiniMax (Hailuo 02 Pro).
Interestingly, in the model updates in June and August this year, whether it is text – to – video or image – to – video, the rankings of “small companies” like Aish Technology are actually better than those of ByteDance.
This overtaking is actually a manifestation of the “divergence” in the development of the video generation track between large and small AI companies.
[1]
Currently, companies like Aish Technology tend to focus on product – oriented and tool – based thinking.
This choice is related to their status as startups. Due to limited core resources such as funds and talents, these companies can only focus on model products, continuously refine AI capabilities, and concentrate on improving the speed, quality, and instruction – following optimization of video generation.
According to Wang Changhu, the founder of Aish Technology, the current focus of Aish is on ordinary people without video production skills. He compares the current product to Canva in the field of video generation, corresponding to Photoshop and Figma.
From this perspective, what companies like Aish Technology want to do seems to be a super AI tool targeting C – end users.
Referring to the experience of the past Internet era, tool – based APPs are undoubtedly a difficult business model. For example, apps like Moji Weather, WiFi Master Key, Wandoujia, and 91 Mobile Assistant were all very popular in the early stage, but had little moat in the later stage of the mobile Internet.
Either their core functions were directly integrated by mobile phone manufacturers, or their business models were single with limited ceilings, making it difficult to convert a large number of users into effective revenue…
Of course, AI video generation also has the potential to become a content community. However, there is still a long way to go technically to truly generate a blockbuster with just one sentence.
In addition, after personally experiencing several AI video generation platforms, the Internet Jianghu team also found some problems: In a video community full of AI – generated content, the problems of homogenization and aesthetic fatigue are particularly obvious. “After scrolling through a dozen videos, you just don’t want to continue, feeling bored.”
At the same time, since everyone is focused on technology, the entire community lacks a “human touch”. Without the appreciation and evaluation of third – party onlookers, content creators do not receive much material and spiritual incentives. This is why many people transfer their AI – generated videos to platforms like Douyin, Kuaishou, and Bilibili.
The AI content production platforms and content sharing platforms seem to be spontaneously separated by ordinary users.
[2]
In contrast, as “veterans” of the Internet era, companies like Google, ByteDance, and Kuaishou have a more comprehensive approach in the AI video generation track. They not only focus on product performance but also pay more attention to ecosystem layout.
On the one hand, they seek synergies with their existing businesses. For example, Google’s latest video generation model, Veo 3.0, is not launched as an independent product but is deeply integrated into an AI movie – making tool called Flow.
Flow also integrates two major models, Imagen (text – to – image) and Gemini (general – purpose), aiming to enable users to achieve seamless AI creation from text story output to shot control and video generation.
Similarly, from the Doubao large – model to Jimeng, Jianying, and Seedream 4.0, and then to Douyin, ByteDance has also built a closed – loop AI ecosystem around content production, distribution, consumption, and commercialization…
On the other hand, they explore the independent potential of AI video generation business.
The most typical example is Kuaishou Keling, which has shown great potential in commercial scenarios such as game production and professional film and television production.
For example, in June this year, Keling AI co – produced the world’s first AI – generated unit story collection, “New World Loading”, with Kuaishou Xingmang Short Dramas, which has accumulated nearly 200 million views globally.
Another example is that at the earnings conference in August, Kuaishou revealed that in the second quarter, Keling AI reached a deep – level cooperation with NetEase Games’ popular mobile game “Yi Shui Han”, integrating AI video generation capabilities into the game to enrich its social gameplay.
In this context, Kuaishou Keling has become one of the few companies in the industry that dares to disclose its AI performance.
According to Tianyancha APP, in the second – quarter financial report, Keling AI under Kuaishou achieved a single – quarter revenue of over 250 million RMB, a 66.67% quarter – on – quarter increase from 150 million RMB in the first quarter. Correspondingly, its annual revenue target for 2025 has also doubled compared to the initial setting, and its investment in inference computing power has also doubled, with a positive gross profit margin…
However, it’s not time to celebrate in the video generation industry yet.
The reason is that the pattern of the AI video generation industry is not stable enough, and there are still many uncertainties.
Firstly, in terms of technology, the capabilities of major video generation models are still in a period of rapid evolution and are far from the ideal performance. So, who can guarantee to stay ahead in the subsequent AI competition?
Secondly, there are potential threats from competitors. Many powerful players in the field of large AI models have not fully entered the game yet. For example, Baidu’s CEO Robin Li was initially reluctant to follow Sora because the commercialization of AI video generation might take five or even ten years. But now, everything has accelerated. Will Baidu, which owns iQiyi, further increase its investment?
In addition, Tencent is also a major concern. Although it has not made many moves yet, it also waited for a long time in the field of large language models. It wasn’t until DeepSeek emerged that Tencent quickly followed up and integrated its model into WeChat, reaping the biggest rewards.
Therefore, for both large and small companies, the AI video generation industry has not reached its end…
From Advertising Companies to E – commerce Companies, and Then to AI Companies
From the recent sharp rise in technology stocks and the intensive actions of major technology giants to increase their investment in AI, it is not difficult to find that:
For Internet companies, the inevitable path in the future is to become a real AI company.
This AI company does not simply mean using AI to empower advertising business or improve the matching efficiency of e – commerce, but the proportion of AI business is increasing.
Take Baidu, which had the sharpest rise recently. The capital market’s recognition of it may be driven by the explosion of its Kunlun chip business and the breakthrough in the proportion of AI and other businesses.
In the second quarter of this year, Baidu’s online marketing revenue was 16.2 billion RMB, a 15% year – on – year decrease; its non – online marketing revenue was 10 billion RMB, a 34% year – on – year increase, mainly driven by the growth of its intelligent cloud business.
If we include iQiyi, the revenue proportion of the latter two businesses exceeds 50%…
In addition, Tencent, which used to connect everything through WeChat, suddenly announced the full opening of its AI capabilities last week to help “useful AI” accelerate its implementation in various industries. Overall, it is moving towards the 2.0 era of AI connecting everything.
In fact, to some extent, the entire Internet industry has reached a critical point of evolving into the 2.0 era.
Since September this year, Internet companies such as Alibaba have been increasing their bond financing and investment in AI, preparing for the future AI competition.
Overseas, recently, NVIDIA and OpenAI announced a strategic partnership, investing tens of billions of dollars to deploy 10 gigawatts of computing power. In the second quarter of this year, Google’s cloud revenue increased by 32% year – on – year. Against the background of a capital expenditure of $22.446 billion, it also raised its annual capital expenditure by $10 billion to $85 billion.
The same reaction can be seen from other technology giants such as Microsoft, Meta, and Amazon.
All these indicate that AI is the ticket to the next era.
In this context, in the video generation track, Keling AI, which has taken the lead, is particularly important for Kuaishou.
On the one hand, it is like the vanguard of Kuaishou’s AI strategy and the second growth curve in the future. It is as important to Kuaishou as Quark is to Alibaba, representing the most difficult breakthrough in the user application layer.
It is foreseeable that with the advantages of having technology (capital), ecosystem, and commercialization in place first, Keling’s revenue flywheel will spin faster with the popularization of the AI era and the evolution of video generation technology.
After all, the former brings a wide range of users, and the latter is the key to deeply binding users. Since both are in a period of rapid growth, the “cake” of AI video generation will naturally grow larger.
Therefore, Kuaishou’s upgrade of Keling AI may just be the beginning. In the future, Keling’s strategic importance will continue to increase, which is determined by the trend of the AI era.
On the other hand, Keling can also be regarded as Kuaishou’s defensive moat.
Why is that?
In the traditional model, the platform is a “marketplace of content”, and its core task is to attract more sellers (content creators) and help them find buyers (viewers).
The maturity of AI video generation may mean a fundamental change in the way of obtaining “content sources”.
At this time, the platform cannot just be a marketplace. It also needs to have the ability of “self – production” or even “providing advanced production tools”. If users’ consumption preferences shift to new – type AI – driven videos (such as highly personalized interactive narratives) in the future, and the platform lacks such core production capabilities, will its ecological vitality and user stickiness face severe challenges?
This logic is similar to the impact of self – driving technology on ride – hailing platforms. The old model of ride – hailing platforms was to integrate scattered driver resources, while self – driving technology may turn the supply side into highly concentrated technology providers.
If ride – hailing platforms are absent in this technological change, they may eventually become mere traffic entrances, with their value greatly squeezed.
Therefore, in terms of strategic importance, both Kuaishou Keling and Didi’s self – driving division must go all out.
Just like in the past Weibo war, many Internet giants such as Tencent, NetEase, and Sohu were all involved in the “Weibo” business, but why did Sina end up being the winner?
An important reason is that for companies like Tencent, Weibo was just an experimental business. It was nice to have, but its absence would not affect their fundamentals.
But for Sina, Weibo was almost its only business at that time, so it had to go all out to succeed.
This may also explain the recent change of leadership at Keling AI.
As those who have experienced battles know, changing defenses is inevitable during major battles. Only by putting the right people in the right positions can we avoid failures at critical moments.
So far, video generation by large models can be seen as a key step in promoting the Internet to enter the 2.0 era.
Only by taking this step can AI gradually evolve from a technical tool to the underlying grammar of a new content world, giving rise to a new paradigm of human – machine co – creation.
The end of this transformation is not to replace anyone, but to open up a more imaginative new content world.
To this day, there are still large blank areas on the maps of the explorers…
This article is from the WeChat official account