Kuailing is merely the next target for Douyin's Dream.

Written by | Lan Dong Business, Zhao Weiwei

Zhang Nan, who stepped down as CEO of Douyin earlier this year, should have had a fulfilling 2024.

In the four-story auditorium of the Central Academy of Fine Arts' School of Design, Zhang Nan, a former art student, sat in the center, smiling as she posed for a group photo with a group of college students. She was one of the judges at an AI Interactive Innovation Competition. The event was organized by Jianying's AI product, Dream, which Zhang Nan currently oversees and is seen as a potential game-changer for Douyin.

A few days later, Zhang Nan publicly described Dream as a camera for the "world of imagination," whereas Douyin is a camera for the "real world."

To create a world of imagination, innovation often stems from young minds. Douyin's success is inseparable from young university students, who were the focus of Douyin's early operations team. Various classic challenge activities spread widely through the enthusiastic participation of these young creators. Unlike Douyin's early days, which were constrained by limited funds and relied on emotional appeals, the first prize for the Dream AI Interactive Innovation Competition is now a cash award of RMB 100,000.

The winner, Zhao Chunxiang, is not a student but a young independent developer who previously gained recognition for a diet-tracking product called "Stomach Book." The award-winning entry was an AI video generation solution with precise control over camera shots in the UI/UX. In a two-minute demo, by importing a classic scene from the movie "Cinema Paradiso," users can generate a video with push-pull shots, close-ups, and blooming flowers outside the window, all created by AI.

Three months before Dream's launch, Kuaishou's Kuailing had already become the first in China to introduce a large-scale video generation model. They too have not overlooked young university students. Kuailing co-hosted an AI Creation Contest with institutions like the China Academy of Art. The three first-prize winners, focusing on life, advertising, and free-form themes, each received a prize of RMB 36,666.

Sora opened the door to AI modeling the real world, while Dream and Kuailing are following in the footsteps of Douyin and Kuaishou, investing heavily to achieve remarkable results.

The bigger competition is looming in 2025. A research memo circulating online about ByteDance's AI video generation products states, "ByteDance hopes to leverage AI capabilities within its ecosystem and believes that next year, each ecosystem will form a closed loop of competition. By May Day next year, the Kouzi Intelligent Agent Platform, Doubao, Douyin, and B-end capabilities will form an interconnected ecosystem, with more manifestations and usage scenarios for text-to-video generation."

In 2025, in markets such as e-commerce advertising and short dramas, Douyin Dream may clash head-on with Kuaishou Kuailing.

A strong start but a weak finish

"A strong start but a weak finish" versus "a weak start but a strong finish" is currently the biggest difference between Douyin Dream and Kuaishou Kuailing.

The latest QuestMobile data shows that on the day of its launch, Dream sparked heated discussions on the Douyin platform, and Kuailing AI also reached a peak of popularity on the Kuaishou platform through continuous accumulation of heat. However, a notable difference is that Dream's content interaction volume peaked within a month and then declined, while Kuailing's content interaction volume gradually increased over the same period.

This disparity is due to a combination of factors such as promotion strategies, user experience, and market competition. A direct reason may be that Dream's product launch was delayed, and the experience fell short of expectations. After in-depth use, users can easily distinguish the quality and stability of generated content compared to similar products. Even though Dream gained significant pre-launch hype, it is still less popular than Kuailing.

This does not mean that Dream is absolutely behind. Users who deeply engage with AI products believe that to effectively use current domestic AI products, one cannot rely solely on a single product. Especially in text-to-video creation, users often choose to use Dream for initial text-to-image generation and then Kuailing for image-to-video creation because "Dream's AI-generated images are superior."

ByteDance's research memo also highlights a significant gap between Dream and Kuailing. Dream has daily active users between 200,000 and 220,000, 70% of whom are individuals or small MCN studios, with relatively few large enterprises. There are approximately 25,000 paying users, with an average monthly subscription fee of around RMB 50. In the same period, Kuailing served over 5 million users, with over 2 million paying users and cumulative payments totaling approximately ten million RMB.

It is difficult to verify the authenticity of such research memos. ByteDance has previously warned investors about the "Doubao Concept Stock" boom in the secondary market to avoid unnecessary investment losses.

Kuailing's "weak start but strong finish" is due to several factors. On the one hand, the video generation large model itself is more stable, giving Kuailing a first-mover advantage. On the other hand, its marketing strategy has been successful. As Lan Dong Business mentioned in "Kuaishou Kuailing Puts Pressure on Douyin Jianying," Kuaishou successfully created a buzz in the overseas market by having tech influencers test Kuailing-generated content, leading to a "made-in-China, sold globally" strategy.

Six months later, Kuailing still garners far more attention in the overseas market than Dream, with 67 times more followers on social media platform X.

On the day Zhang Nan appeared at the Volcano Engine Conference to announce the latest news about Dream, Kuaishou upgraded its Kuailing large model, claiming a 195% improvement in overall effectiveness compared to the previous 1.5 model in internal evaluations. During the third-quarter earnings call a month earlier, founder Cheng Yixiao also expressed optimism about Kuailing, noting that Kuailing AI's monthly commercial revenue exceeded ten million RMB and expressing confidence in rapid revenue growth next year.

Dream, which had a "strong start but a weak finish," aims to become a new way of creation and experience. According to the research memo, Dream has no clear commercial return target for next year but aims to establish a business model. "Profitability will come later," and Dream's focus next year will be on product implementations such as collaborations with media and film production.

Kuaishou will once again face an encirclement battle

Douyin was not the first to enter the short video market but surpassed Kuaishou in 2018 to become the leader. Zhang Nan once summarized four key factors for Douyin's rise: full-screen HD, music, special effects filters, and personalized algorithm recommendation technology.

Now, with Dream versus Kuailing, can history repeat itself, with Douyin overtaking Kuaishou once again?

Dream is currently just one product in ByteDance's multimodal large model application layer, belonging to Douyin's Jianying team and supported by ByteDance Cloud Services' Volcano Engine. On the Volcano Engine's model plaza, ByteDance offers 20 large model products across text, speech, and vision. Additionally, Volcano Ark provides products from Moon's Dark Side and Zhipu AI.

AI could become ByteDance's next core business pillar. In contrast, it is difficult to find displays of related large model business applications on Kuaishou's official Magnetic Engine website.

ByteDance's aggressive stance in the large model field was already evident through its C-end product Doubao earlier this year. In September, mobile data research firm Sensor Tower released a global AI app report showing that ChatGPT was the most downloaded AI app worldwide from January to August, with Google's Gemini ranking fourth and ByteDance's Doubao ranking fifth, making it the only Chinese product on the list.

This success is inseparable from Douyin's abundant traffic, advertising, and promotional support. Doubao and Kimi competed fiercely in the advertising market this year. According to ad intelligence analysis platform AppGrowing, Doubao's advertising investment amounted to nearly RMB 18 million in April and May, surging to RMB 124 million in early June, with restrictions placed on large model advertising, including Kimi, within the Douyin platform.

"In terms of marketing budget, Dream's budget will increase in December, and in the first quarter of next year, especially around the Spring Festival, it will reach hundreds of millions of RMB," according to the ByteDance research memo. Besides marketing, ByteDance's chip-level reserves cannot be underestimated. The Financial Times reported that ByteDance has purchased about 230,000 NVIDIA chips, becoming the largest Chinese buyer of NVIDIA AI chips. The Information also reported in September that ByteDance ordered over 200,000 NVIDIA H20s this year.

Given Doubao's leading position in the domestic large model market, the future focus will be on how Douyin and Doubao can synergize with Dream. This also means that Kuaishou's standout product Kuailing will face an encirclement battle from ByteDance's large models.

In September this year, ByteDance released two text-to-video tools, PixelDance and Seaweed, to compete with OpenAI's Sora. Dream AI has already integrated with Doubao, with support from the more capable PixelDance. Officially, it can generate high-quality, two-minute-long 1080p resolution videos, excelling at depicting complex movements and interactions between objects.

Currently, whether it's Douyin or Kuaishou, the primary application scenarios for AI-generated videos are similar. Besides charging C-end users, B-end scenarios include serving the film and television production and post-production markets, such as short dramas, as well as serving advertising and e-commerce content marketing, like generating different images for product displays.

At the Volcano Engine Conference, Zhang Nan showcased two AI short films created by Dream users, one of which was the sci-fi short drama "Awakening" launched in July this year, which received over 400,000 likes on Douyin in a single day. During the same period, Kuaishou's Kuailing also produced "The Miraculous Splendor of Mountains and Seas: Sailing Against the Waves," both of which were trial productions.

However, AI-generated videos are currently auxiliary to film and television production, still at a small scale. To complete large-scale post-production, both Dream and Kuailing are advancing along similar Dit architectures (a diffusion model combining Transformer architecture for image and video generation tasks), with a long way to go and commercialization still premature.

When Google Beats Sora

After OpenAI's Sora was opened for use, a series of generated videos did not meet external expectations. Recently, Google's newly released video generator Veo2 demonstrated more impressive performance through a series of tests.

Particularly notable is a scene of slicing a tomato, where Google's Veo2 cleanly slices the tomato while avoiding fingers, whereas the knife in the Sora video slices through the hand, making Sora the butt of jokes again and leading industry insiders to believe that Sora is more motion-oriented, while Veo2 focuses more on physical accuracy.

Some AI industry insiders believe that Google's ability to surpass Sora is not only due to identifying Sora's weakness in physical accuracy but also to using YouTube to train its AI models.

ByteDance's technical team is well aware of Sora's weakness in physical accuracy. In November, the Doubao large model team published a paper titled "How Far is Video Generation from World Model: A Physical Law Perspective," exploring whether video generation models can observe the interrelationships between objects and extract a set of stable physical laws.

"Visual ambiguity can lead to significant errors in fine-grained physical modeling, and relying solely on video representation is insufficient for accurate physical modeling," the paper argues, noting that video generation models still face challenges in becoming accurate world models.

The two authors researching this direction are both very young, one born after 1995 and the other after 2000. Like Dream and Kuailing, which require the participation of young art students to create a world of imagination, the technical foundation for this AI imagination world also comes from young minds. The two authors spent eight months finding a gateway to the world model.

Finding a bottleneck takes eight months, but breaking it may take even longer.

When will Douyin truly become a dream? The research memo outlines three main paths for ByteDance's AI development next year: the Doubao family ecosystem, comprehensive AI integration into products like Douyin, and multimodal and world models including Dream. The multimodal path is a priority, with "unlimited support and investment because it is a crucial node in the transformation, and larger losses are acceptable."

When Google beats Sora, it heralds the breaking of the model myth created by OpenAI. And Kuaishou's Kuailing is merely the next target for Douyin Dream.