On May 13, Dream AI topped the Apple App Store free app download list for the first time, surpassing the perennial bean bag. Behind this victory is not only the embodiment of the technical strength of Instant Dream AI, but also the result of Byte’s layout and strategic adjustment in the field of AI. This article will provide an in-depth analysis of the rise of Dream AI, as well as Byte’s strategic layout and future challenges in the field of AI video generation.
Since May,Byte has further increased its emphasis on its video generation application Dream AI.One of the intuitive manifestations is that Dream AI has risen rapidly in the ranking of free app downloads on Apple’s App Store in China.
Since May 13, Dream AI has surpassed Byte’s two top two applications that dominate the list all year round, and reached the top for the first time.
Before Byte, another short video platform, Kuaishou, attached great importance to AI video generation was first reflected in its organizational structure.
Recently, Kuaishou has upgraded Keling AI to a first-level business unit alongside the main station, commercialization, e-commerce, etc. by establishing the Keling AI Division. Before Kuaishou, Byte also upgraded its AI application department Flow, which was established after 2023, and Seed, the model research and development department, to first-level departments alongside Douyin, Feishu, Volcano Engine, etc.
nowInternal and external pressure has been given to Zhang Nan, the former CEO of Douyin.Looking around Byte’s newly established AI business, almost only Zhang Nan’s Dream AI is left, which has not yet ushered in business upgrade opportunities. In addition to Flow, Dream AI is another main way for Byte to explore AI applications. In a sense, the success or failure of Dream AI is directly related to whether Zhang Nan can return to the core of byte power.
When announcing his resignation as CEO of Douyin Group in February last year, Zhang Nan said that he would focus on the development of Jianying in the future. As a video editing software launched by Douyin in May 2019,Zhang Nan, who has rotated from the core business Douyin to the relatively edge business, has been labeled as “exiled” by the outside world.
But Zhang Nan’s ambition is to bet on AI and achieve internal re-entrepreneurship. OpenAI became the catalyst for Zhang Nan’s determination.
How can product managers do a good job in B-end digitalization?
All walks of life have taken advantage of the ride-hail of digital transformation and achieved the rapid development of the industry. Since B-end products are products that provide services for enterprises, how should enterprises ride the digital ride?
View details >
In 2022, the release of OpenAI’s Wenshengtu model DALL-E 2 allowed Zhang Nan to intuitively feel the disruptive power of AI image generation to traditional content creation methods for the first time, which is not only an opportunity for Byte CEO Liang Rubo to “create a new creative platform”, but also one of the reasons for Zhang Nan’s transfer to cut.
Three months after announcing his resignation, in May last year, Zhang Nan handed over his first assignment after transferring – positioning the AI creation community and supporting image and video generationThe latter has become an opportunity for Zhang Nan to recreate the next Douyin.
According to the LatePost,At present, Zhang Nan spends more energy on Dream AI, and the editing is handed over to his direct subordinates.
QuestMobile data shows thatAs of March 2025, the number of monthly active users of Dream AI has reached 8.93 million, while that of Keling AI is 1.8 million。 The gap in the amount of data between the two products reflects the different strategies of Byte and Kuaishou for video generation products to a certain extent: Dream AI is more inclined to develop ordinary users and prioritizes user expansion; Keling AI pays more attention to the penetration of professional users and prioritizes the pursuit of revenue growth.
However, for Jimeng AI, which is on the road to pursuing user expansion, its monthly active activity of less than 10 million is still “108,000 miles” away from the scale effect that Zhang Nan wants to achieve. After all, compared with another star AI application Doubao, which has exceeded 100 million monthly users in the same period.
If he wants to return to the core of Byte, Zhang Nan must at least prove that Dream AI has the potential to not lose to Doubao.
01
In this competition in the field of video generation,Zhang Nan was expected to lead Byte to create a lead.
After the debut of Sora in February last year, a number of domestic AI manufacturers began to focus on creating a “domestic version of Sora”. As early as one month before the release of Keling AI, in May last year, Dream AI launched a video generation function at the beginning of its launch. However, because the generation effect is average, the dream AI has not ushered in a circle-breaking effect.
Zhang Nan’s missed opportunity fell on Kuaishou. In June last year, the Kuaishou Keling large model was launched, becoming the world’s first video model to adopt the DiT route in line with Sora and open directly to ordinary users, making Keling AI gain millions of users in a short period of time.
It was not until September last year that Byte gradually stabilized its formation with the release of PixelDance and Seaweed’s two large video generation models. Two months later, in November of the same year, Dream AI won the first qualification for the video model and became the main entrance to the public to experience it for free.
After nearly a year of development,Dream AI and Keling AI are gradually converging from video model performance to product functions. One of the intuitive manifestations of this convergence is the access to DeepSeek R1.
In March, under the premise that neither Douyin nor Kuaishou announced their access to DeepSeek, Dream AI and Keling AI successively chose to access DeepSeek to assist users in generating inspiration words for images and videos.
Caption: Keling AI product page
In terms of product functions, both Dream AI and Keling AI can realize image generation, Wensheng video, picture generation, digital humans, etc. Specific to the field of video generation,Both products are available in both free and membership versions.In the free version, the video duration includes 5s and 10s, and it costs 2 points to generate a 1s video (in Keling AI, points are called inspiration values).
In addition to using the free points given by daily login, there are two ways for users to obtain points: one is to recharge directly, and 1 yuan can buy 10 points; The second is to open a membership, that is, Dream AI will continue to subscribe to 69 yuan per month, with 1080 points per month, and Keling AI will continue to subscribe to 66 yuan per month, with 660 inspiration points per month.
In terms of price alone, Byte has once again played its cost-effective strategy in the field of large models.It is worth mentioning that after opening a membership, in addition to enjoying more points rewards, users can also remove brand watermarks from generated works and extend the duration of generated videos.
Caption: Dream AI product page
Among the major technology companies, Byte and Kuaishou have also become the two most radical companies in making video generation products. After Sora set off a wave, Baidu founder Robin Li made it clear that he “will not chase Sora”.
This is also related to the fact that the main short video business of the two companies is more susceptible to video generation.
In 2019, the emergence of video editing tools such as clipping helped the content ecology of short video platforms shift from PGC (content produced by professional institutions) to UGC (user-generated content), greatly lowering the threshold for user creation.
Nowadays, with the emergence of AI-generated video products, it is expected to further lower the threshold for every ordinary person to create videos, thereby driving a new explosion in the form and quantity of content creation.
In addition, specific to the business level of short video platforms, image/video generation can also help improve the commercialization needs of platforms, such as helping Byte and Kuaishou advertisers produce videos at low cost and conveniently.
02
Dream AI, which has convergent product functions and a higher number of monthly active users, is not waiting for the opportunity to upgrade its business like Keling AI, and one of the reasons is the AI development strategy of Byte and Kuaishou.
Unlike Kuaishou, which bets all its AI ambitions on visual generation, Byte chose to make an all-out attack.In addition to the main visual generation of dream AI, there are also AI assistant products such as Doubao, entertainment companion products cat boxes, and AI education products such as Doubao Love Learning. I hope to sweep away all the possibilities of the AI track.
But the price of this isIt is inevitable that the teams will fall into an internal horse racing mechanism.
Before Zhang Nan’s transfer to Jianying in February last year, Byte already had three teams around the development and application of large model technology: Zhu Wenjia, head of TikTok product technology in Singapore, Xiang Liang’s team, head of Data-AML (Data-Application Machine Learning) in Seattle, and Li Hang, director of Byte AI Lab (artificial intelligence laboratory) in Beijing.
At that time, a person familiar with the matter told Alphabet List that the image team where Jianying was stationed was stationed in Xiamen, which also meant that with Zhang Nan’s entry into the AI large model battle, in China alone, Byte had two teams that competed with each other in cooperation.
Such a horse racing mechanism is directly reflected in the dream AI.Unlike Keling AI, which iteratively upgrades on a single model, Dream AI accesses video models developed by different teams.
At present, the five video generation models available to users for Dream AI come from four different R&D teams: the video version 1.2 is from the AI lab; Video P2.0 Pro version from PixelDance; Video S2.0 Pro and Video S2.0, both from Seaweed aipha; Video version 3.0 is from Seedance 1.0.
In addition to the temporary difficulty of business upgrade opportunities due to the horse racing mechanism,Compared with the video generation in the niche track of Dream AI, conversation applications are obviously a challenge that Byte needs to deal with at present.
Before DeepSeek became popular, Doubao had successfully ranked among the largest AI assistant products in China. However, since the beginning of the year, with the dividends obtained by being the first to access DeepSeek, Tencent has only used 1/10 of the time and a much smaller investment budget to reduce the user scale of Yuanbao to about 1/4 of Doubao (as of March, Doubao has 116 million monthly active users and Tencent Yuanbao has 41 million).
In addition to Tencent, technology giants, including Alibaba, have also rekindled their interest in AI to C. With many rivals, the growth momentum of bean bags is being curbed. According to QuestMobile data, in terms of the compound growth rate of weekly active users in the first quarter, DeepSeek was 58%, Tencent Yuanbao was 23%, and Doubao was only 3.39%.
03
Although Dream AI has not yet become Byte’s first-level business, butZhang Nan still holds the cards of breaking the game in his hands, and the AI community is one of them.
With more than 100 million monthly active bean bags, it has not really realized the community function, and its “creation” page can be seen in the application, and users can also share creative cases, but they cannot interact with other users’ messages.
At present, in order to enhance user stickiness, AI products have gradually become a consensus to be a community. Recently, there have been reports that Kimi is testing the community function in the product and OpenAI is being tested on social networks.
At the beginning of the establishment of Dream AI, Zhang Nan set the goal of an AI creation community for it, and hoped to use it to create an AI version of Douyin.
Even, according to the “Intelligent Emergence” news, in December last year, Byte’s management also judged that AI dialogue products such as Doubao may only be the “intermediate state” of AI products, and the long-term more ideal product form is likely to require a more visual user experience and a lower threshold for user use. Therefore, Byte has raised the product priority of Dream AI.
However, in the ever-changing field of AI, not only is model technology iterating rapidly, but the industry’s cognition of AI products is also constantly updated. Judging from Byte’s past product development experience, there is no complete community and stronger generation ability of bean bags, and there may not be one in the future.
Referring to the lessons of the past of the annexation of volcano videos and watermelon videos after winning the Douyin horse race, once the bean bags are all the way to the dust in terms of scale effect, they may not repeat a new wave of annexation.
And such signs have already occurred. In mid-to-late April, “LatePost” reported that Liang Chenqi, the head of Byte Cat Box, left, and the cat box was taken over by Xinghui product leader Nishihara (pseudonym), while the Star Paint team planned to merge into Doubao. This also means that Star Painting, which originally belonged to Flow’s AI products, has now become a part of Doubao in terms of organizational structure.
Even if Dream AI will not be merged by Doubao, with the iteration of multi-modal large model technology, using one model to solve all user needs will shine from theory to reality. If Doubao also integrates multimodal capabilities, it may not directly integrate stronger generation capabilities into it.
At present, by selecting the “Photo Move” function in the bean bag, users can already complete the operation of Wensheng Video and Tusheng Video. However, compared with Dream AI, the video generation function in Doubao only supports 5s duration for the time being, and does not support more functions such as resolution and video extension.
In the face of the dual pressure of user scale and technology iteration, whether Dream AI can become a new first-level business unit of Byte or a sub-function of bean bags depends on Zhang Nan’s performance.
