Kuaishou released its Q2 2024 earnings report yesterday: Revenue increased by 11.7% YoY, and adjusted net profit increased by 73.7% YoY; both significantly exceeding market consensus expectations. Kuaishou's MAU (Monthly Active Users), DAU (Daily Active Users), and user engagement time continue to grow, whether it's user data or revenue and profit figures—this is undoubtedly a financial report that leaves no room for criticism in the capital market.

However, compared to purely financial figures, I am most concerned about the performance of generative AI, which will determine the future of all internet content platforms, including Kuaishou. I've joined several AIGC exchange groups, whose members are primarily practitioners in AIGC technology or product directions, either working for major internet companies or starting their own businesses. Since June this year, "Keling" has become one of the hottest keywords in these groups. Here's a chronological summary:

June 7th: "Kuaishou's Keling is open for beta testing!" "So cool?" "Cool!" (This formation continued for over ten posts) "The only drawback is that it can't generate videos from images." (This drawback was soon addressed) June 27th: "Looking for someone who knows Keling, is there a commercial API available? We have a lot of demand." (Unfortunately, there wasn't at the time, and the asker was quite disappointed) July 14th: "I used Keling to create an animated story from start to finish. I think entrusting all concept animations to Keling is very feasible." (Spoken by a leading game artist in China) "Impressive, boss!" (A group member's reply) August 6th: "Keling is not only the best-performing video generation model in China for both text-to-video and image-to-video, but it's also the only truly mature one. The others are all half-baked and can't compare." (Note: This was two months after Keling's release)

To this day, Keling remains one of the few video generation models worldwide that leads in performance and truly has practical value. As a moderately successful video creator, while I haven't fully embraced AIGC yet, I've been following its progress. Just yesterday afternoon, I used Keling to generate a video of cats playing in a cardboard box and compared it with real videos of my own cats. For me, Keling's biggest drawback at present is that it can only generate 10-second videos at a time. Of course, for creators who are already proficient in shot division, 10 seconds can meet their needs. But for me at this stage, if future upgrades to Keling can support generating longer videos at once, I would be more than happy to rely on AI for my daily video creation.

Anyone with even a little video creation experience should realize the revolutionary significance of AIGC tools like Keling. The two foundations of video creation are copywriting and materials, with the basic creative process involving writing the copy first and then filming or sourcing materials based on it. Seamlessly integrating materials and copywriting is a formidable task! Major video platforms provide creators with vast material libraries, but that's far from sufficient. While "institutionalized" creators (e.g., MCNs) can invest in shooting or procuring higher-quality materials that better align with their copywriting, what about the vast majority of PUGC creators? They are the backbone of short video platforms' content ecosystems, and without them, any platform's content ecology would become stagnant.

From this perspective, Keling, an AI tool, and Kuaishou, the platform, are a perfect match: Kuaishou is renowned for its "full coverage" of users and content, with its slogan being "Embrace Every Kind of Life." The imaginative ideas of millions of diverse users are the soul of Kuaishou's content ecosystem. Over the years, countless viral memes have originated from Kuaishou users' flashes of inspiration, which were then imitated by other platforms, ultimately "going viral." As one of the earliest short video platforms in China, Kuaishou once "delegated" the channels or say, the right to speak, of video creation to millions of creators; now, through Keling, it is "delegating" efficient video creation tools to millions of ordinary people.

I believe that for video creators who do not primarily publish on Kuaishou, a common practice over the past two months has been to download Kuaishou's official video editing tool, apply for Keling beta access, and experience the deeply integrated Keling creation tool. Of course, Keling now offers a web version that can be used directly, but the experience is better when combined with Kuaishou's video editing tool. After some time using it, some of them will surely seriously consider whether to make Kuaishou one of their primary creative platforms. Take myself as an example; although I've been creating content on Kuaishou for three years, before trying Keling, it was just one of many platforms where I published content; after trying Keling, its importance in my mind has grown, and it may become my most important publishing platform. This is a microcosm of AIGC's role in expanding Kuaishou's content ecosystem! This expansion is comprehensive and continuous, and its impact is just beginning to manifest.

The above discussion focused on content creation. In reality, AIGC's revolutionary impact is more evident at the commercialization level. This has been fully demonstrated in Silicon Valley, the birthplace of the AIGC trend. For example, in its recently released Q2 earnings report, Google proudly announced that "AI-coproduced videos" between Verizon (North America's largest telecommunications operator) and YouTube creators can display ad videos of varying lengths and focuses to different audiences; in this series of videos, Verizon's cost per acquisition (CPA) decreased by 15%, and conversion rates increased by 38%. Both Google and Meta (Facebook) management are optimistic that once generative AI matures, advertisers may only need to provide targets and budgets to the platform, with AI handling the entire process from creativity to execution. This day is not far off!

Kuaishou's Q2 earnings report also prominently featured the theme of "AI-driven commercialization": "Our AI matrix has been seamlessly integrated into multiple business scenarios, supporting content creation, content recommendation and understanding, and user interaction. In June 2024, the average daily consumption of AIGC marketing materials from marketing clients reached RMB 20 million." During the earnings call, Kuaishou CEO Cheng Yixiao further noted, "We are paying more attention to the application of large models in commercial scenarios. Video and live stream script generation and advertising lead customer service, built on the foundation of the Kuaiyi large language model, combined with digital human technology, help commercial advertisers produce high-quality video and live stream content at low cost and improve lead conversion efficiency."

Cheng Yixiao has described it clearly: Kuaiyi (the foundational large language model), Ketu (the image generation model), and Keling (the video generation model) together constitute an integrated AIGC solution covering the entire advertising creation process. Merchants can use Kuaiyi to generate copywriting, then Ketu to generate images based on that copywriting, and finally Keling to complete the task of generating videos from images—this integrated generation path will make the final content more consistent and aligned with expectations. And that's not all; in the content distribution stage, both regular and commercial content like ads can be empowered by the recommendation large model to make recommendations more in line with users' preferences; for merchants, this translates to higher ad conversion rates!

See? The concept proposed by Google and Meta, where advertisers only need to provide targets and budgets, and AI handles the entire process, is already taking shape and being implemented at Kuaishou. If advertisers find that Kuaishou's "AI-integrated solution" can bring lower costs and higher efficiency, they will undoubtedly prefer to advertise on Kuaishou and even shift their entire business towards the platform. To date, when discussing Kuaishou's unique value to brands and merchants, the most frequently mentioned aspects include its "combined public and private domain" traffic allocation mechanism, its comprehensive coverage of users at all levels (especially in lower-tier markets), and the diversity of its content tone. From now on, AI may become an equally important reason and even Kuaishou's core advantage for advertisers. Of course, this process won't happen overnight; it's only just beginning.

At the end of this article, I'd like to pose an intriguing question. Since Keling's launch, many people around me (including myself) have been asking: Why was it Kuaishou that first came up with such a large model?

In the absence of generational differences in technical routes, computing power resources, and training data, the key difference may lie in engineering. Ultimately, 99% of the problems in generative AI research and development are engineering problems, especially those related to properly assembling and managing teams to complete tasks. Don't underestimate this issue; Google's Gemini large model suffered from severe engineering management flaws, leading to continuous delays and even scandals related to its reward model (responsible for aligning with human values) after its release. My personal guess is that Kuaishou efficiently and skillfully completed the engineering tasks at the training level, gaining at least a few months of first-mover advantage over its competitors.

Furthermore, large model technology is still evolving rapidly, and respecting technology is crucial. We cannot approach large model development with the mindset of creating traditional internet products; instead, we need to find a balance between technology and product development. The engineering issues mentioned above are essentially a branch of technical issues. As one of my friends put it, "Kuaishou is very generous to top technical talent. Compared to competitors, Kuaishou has become relatively lean and agile, which is particularly evident in generative AI research and development." By the way, this friend does not work at Kuaishou and is purely making these observations from the sidelines.

In summary, the success of Kuaishou's generative AI tools, including Keling, not only demonstrates the strategic vision of its management but also exemplifies the excellent execution capabilities of the organization. As for how significant this organizational execution advantage is, it's hard to say. But I want to point out that Kuaishou's advantage will persist and expand as long as competitors cannot yet launch tools that rival Keling, gradually leading to a qualitative change from quantitative accumulation. We are witnessing this process unfold!