Tucson's "Content Creation" Future

The once-leading self-driving company, after going through delisting and transformation, has officially announced its entry into the AIGC field. In the future, it might become a competitor to many game companies such as Tencent, NetEase, and miHoYo.

However, the company's former co-founder and former CTO, Hou Xiaodi - a once key figure who fought alongside them, publicly stated: "TuSimple's transformation is like Messi announcing that he's switching to playing basketball. … It's simply impossible for TuSimple to develop an AI large model and game applications. TuSimple should go bankrupt now. Without Hou Xiaodi, there is no future."

This is a plot that even the most carefully crafted business drama would struggle to write. Hou Xiaodi has so dramatically pushed TuSimple Future into the eye of the public opinion storm.

In November, the company's former CTO Hou Xiaodi set up the SaveTuSimple website and called on shareholders to immediately liquidate TuSimple Future and return all the cash to the shareholders. However, at the just-ended TuSimple shareholders' meeting, the plot suddenly reversed.

The majority of shareholders chose to stand on the side of the founder Chen Mo and CEO Lv Cheng, supporting the company's chosen direction of the animation and game business, bringing this internal dispute to an end.

In March 2024, the TuSimple management began to explore the feasibility of the AIGC business.

In August, the board of directors officially approved the transformation into the animation and game field, and at the same time announced that it had obtained the animation film and game authorization cooperation of the "Three-Body Problem" IP.

In December 2024, TuSimple officially announced its name change to CreateAI and simultaneously released several new business progress updates, including obtaining the official authorization of the famous martial arts IP "The Legend of Condor Heroes" to develop a large AAA-level martial arts open-world RPG game; collaborating with the renowned Japanese animation director Shōji Kawamori and the top animation production company Shirogumi Inc. to use AIGC technology to drive creators to achieve more breakthroughs in showcasing the grand cosmic civilization. At the same time, CreateAI also released its first large-scale model product "Ruyi", which is positioned in the open-source image-to-video field.

In just half a year, the TuSimple team completed a major transformation and responded to many people's doubts. While the idealistic Hou Xiaodi is still stubbornly insisting on self-driving entrepreneurship, the realistic Chen Mo led TuSimple to quickly turn around and move towards the animation and game field, which is more easily commercially achievable.

Is TuSimple's this transformation a smooth path filled with flowers or a turbulent undercurrent with hidden reefs? What key capabilities does TuSimple need to possess to become an excellent AIGC and digital content production company?

The "Life-and-Death Game" of AIGC

In fact, since the explosion of large models, it is not uncommon for large companies to transform into or re-invest in AI businesses. TuSimple is just more special due to its delisting and internal turmoil, under the heavy pressure of expectations and pessimism.

For a proven technical team, when doing a "second entrepreneurship", what needs to be worried about is not the technical reserve, but how to transform the technical advantages.

First of all, the key technology of AIGC is not as complex as self-driving. Taking the multi-modal large model as an example, the barriers lie in three aspects: data, computing power, and algorithm.

Computing power is the common bottleneck in the industry, the algorithm is TuSimple's traditional advantage, and the data is "each showing its own strengths".

Since the underlying large model market is already sufficiently segmented, users often focus on the generation level in a certain scene and a certain style. This also implies that the data quality and data processing method of model training are more important than the asset scale.

The closed-loop development capability from data collection, model training to efficient deployment is basically a "standard configuration" in the field of self-driving. It is not difficult to transform and implement a set of mature technical workflows. What TuSimple needs to do is to supplement the industrial strength and improve the understanding and deconstruction ability of vertical data.

In addition, self-driving is a cutting-edge industrial track under the integration of multiple fields, while AIGC still remains in the form of digital application. Whether it is the engineering cycle or the difficulty of technology integration, as a pan-AI field, self-driving will not have too much entry barrier, but is expected to further compress the reasoning cost.

It is not accidental from a technical perspective that TuSimple was able to develop the Ruyi large model in a short period of time. And from a commercial perspective, whether the determination to fight with one's back to the wall can be rewarded will soon have an answer.

In fact, at the end of 2024, whether a company can be AIGCized often does not depend on whether there is a technical genius behind it, but on whether the iteration speed for the target application scenario is "fast, accurate and ruthless" enough.

After the official release of Sora with mixed reviews, a derived consensus is that the time window for entering the AIGC field has not yet closed, at least in the multi-modal fields such as video generation.

At the same time, domestic manufacturers that have quickly become unicorns under the dividend of large models are also inevitably "shaky" in the face of commercialization.

In the past year, AIGC is undoubtedly a big gamble for capital. ByteDance, which started later, has now overtaken others and become the technology giant with the most AI application layouts. In the fierce competition with the big companies, the "Six Tigers" have changed from focusing on commercialization to trying their best to "stay in the game".

It can be said that at this stage, everyone is looking for nails with a hammer. For any manufacturer, doing down-to-earth implementation, finding a profitable direction, and planting large models in the industry is more difficult than simply lowering prices. The position is not that important. Being able to make money from an industry with high growth potential is the solution to break the deadlock of AIGC.

In this environment, TuSimple, which is starting anew, found the application foothold at the first time: ACG.

According to the plan, CreateAI will launch two AVG and SLG games in 2025, and will launch the open-world game of "The Legend of Condor Heroes" and the film-game integration animation film and game under the "Three-Body Problem" IP in 2027.

Based on TuSimple's expectation of making "The Legend of Condor Heroes" into a 3A work comparable to "Black Myth: Wukong", the estimated research and development time of about three years, which is much lower than the average market cycle, basically proves that AIGC will play a key role.

The natural fit between AI generation and the content industry does not require much argument. But what surprises the industry is that TuSimple, which is known for its technology, did not choose to cooperate in development, but insisted on creating a closed-loop ecosystem for its own content while deploying a series of large models and launching basic production tools.

The "Racing Ahead" of Domestic Image-to-Video

When it comes to the future battlefield of AIGC commercialization, technical feasibility is the admission ticket, and the industrial entry point is the real weapon.

Reversing from the time point, that is, after OpenAI released the test version of Sora, TuSimple began to firmly believe in the direction of AI-generated videos, and chose the image-to-video with higher barriers and more in line with the accumulation of self-driving technology.

Especially in large-scale project development, text-to-video can currently only be used for initial creative understanding, while image-to-video can deeply participate in the creative efficiency process by controlling the initial image, model and parameter characteristics.

Therefore, in the film and game industry with heavy digital assets, the video large model is not an intelligent gimmick, but a productivity tool that dynamically expands creative content. This also makes this technology easier to find an anchor point for cost reduction and efficiency increase in the short term.

Globally, the video large model is still in its early stage. Instead, in the domestic market, this time it has not been left behind in the time difference and has achieved a head start in product experience:

In June this year, shortly after the launch of Kuaishou's Keling large model, the image-to-video function was introduced. Subsequently, Douyin launched Jimeng. "Startup large models" such as Vidu and PixVerse also performed well.

The Ruyi large model is based on the DiT architecture and through the underlying infrastructure such as multi-modal data sets, computing, storage, and GPU resource training shared with TuSimple's self-driving business. The currently open-source Mini version can generate a 120-frame/5-second video, and the professionally deployed model planned to be released in the future will reach the 2K HDR image level, and is not inferior to Sora in terms of generation control, motion amplitude, and lens control.

The potential winning point of Ruyi is that almost no other video large model currently has the ability to be directly used for professional-level content production.

Therefore, it is not wrong to say that TuSimple "followed the trend" and chose AIGC, but it did not choose a conventional entry point. The business model of self-developed AI technology & original content distribution itself breaks through the market's inherent assumption of large models for SaaS monetization and content providers.

It may take at least 5-10 years for a fully automated production video large model to appear. But at present, the main pain point in the anime and game industry is not the depletion of content capabilities.

As the fastest-growing sub-sector in the media industry, the global video game industry achieved a total revenue of 196 billion US dollars in 2023, exceeding the combined total of streaming media videos, music, and movie box office.

At the same time, the anime field also maintains a compound annual growth rate of more than 10%. ACG is becoming the most successful global mainstream culture in the content production field in China and even East Asia.

In contrast, after experiencing the capital cold wave, it has become very difficult for high-quality 3A games and animated feature films to both adhere to the pursuit of quality and successfully endure the long production period as before: A Disney animated film takes 4 - 6 years to produce, and 3A games such as "Red Dead Redemption" and "Cyberpunk 2077" have a development cycle of about 7 - 8 years, with the video production cost accounting for 50 - 70%.

Video AIization can replace the "labor-intensive" links of high-quality content production with AI by re-standardizing the production process. CreateAI's goal is to reduce the production cost and development cycle by 70% within five years.

Therefore, TuSimple's entry into the market seems to be a "model war" on the surface, but in fact, it is a blue ocean of digital content efficiency. The ACG - GEN tool it is expected to launch next year is to engineer the capabilities of the Ruyi large model to output a set of self - used and customizable professional - level productivity tools for the market.

This ecological path is not difficult to associate with Epics Games, which is famous for its Unreal Engine. As a game company, it also opens its professional production capabilities to the industry on the basis of self - research and self - distribution.

Currently, CreateAI has open - sourced the Mini version of the Ruyi large model to the Hugging Face community and plans to "quickly enter the circle" by attracting talents from the film and game industry and building a creator ecosystem.

In the technical lead of image - to - video, TuSimple's exertion time is far earlier than the technical maturity period. However, in the future, as the performance of computing power, display and other hardware equipment improves, the industry's demand for the complexity and precision of AI will also increase.

Technical trends such as image - to - video and video - to - video will inevitably come to the center of the stage. At that time, what each manufacturer competes for is the model iteration speed and ecological data capability.

From the current stage, TuSimple has sufficient cash reserves to support the iteration of the large model. At the same time, their new business foothold is also calling to the market: A technology company is not only a content provider.

Rushing Towards Content Manufacturing

If the large model is a torch, then the first batch of industries that "eat the crab of AI" are the sparks of hope.

Not only video generation, but also image, music, 3D modeling, and any professional - level AIGC tool in any细分领域 have the opportunity to redefine creation like Office and Adobe. However, in the general environment of the domestic enterprise service market, it is difficult for those who only make tools to find the most comfortable living space and ceiling.

The direction chosen by TuSimple is the same as the demand of ByteDance's hundreds of thousands of employees for Feishu: To make an advanced large model, it cannot only rely on training, but also on people to use it.

After obtaining the IP authorization of "Three - Body Problem", TuSimple also invited the manga master Shōji Kawamori to lead the production, and by applying the Ruyi large model in Shōji Kawamori's team, the AIGC technology is deeply integrated into the entire animation production ecosystem.

In addition, the founder Chen Mo himself is an animation and martial arts fan for many years. Under the long - term and in - depth contact with the copyright owner, he obtained the authorization to develop the "The Legend of Condor Heroes" IP into an open - world 3A game.

Ten years before the birth of "Black Myth: Wukong", we were all asking the domestic game market a question: Why can't China make 3A games?

The concept of 3A itself represents a long cycle, high budget, and high cost. These three characteristics are positively correlated with the top game quality experience. 3A is the best manifestation of cultural and artistic soft power in the game industry.

Looking back at the development history of Chinese manufacturing, almost all industries have gradually moved from OEM, processing, and hit products to high - precision and sophisticated products. The essential advantage of domestic products lies in the speed of innovation and iteration and the cost reduction and efficiency increase brought by the mature supply chain.

Therefore, in an era where the industrial chain cannot bring cost advantages, content production cannot become "content manufacturing". Although the domestic game market has an excellent growth curve and overseas performance, it is understandable that it pays more attention to survival, hit products and revenue.

The AIGC era is undoubtedly the best moment for the transformation of the domestic cultural content industry.

It is not difficult to see that the reason why TuSimple dares to use the industrialized thinking to "challenge" the traditional content industry is that it aims at the consumers' demand for high - quality content, which is far higher than the content industry's own awareness and ability to improve efficiency.

Alibaba and ByteDance understand the efficiency of Internet office work, so they made DingTalk and Feishu. TuSimple understands the engineering efficiency of intelligent manufacturing. It is practical and feasible to find a balance between the high return, high reputation, high cultural output of large - scale productions and the development cost of several hundred million US dollars.

The disagreement between Hou Xiaodi and the current management of TuSimple is actually more of a divergence in concepts. Netizens believe that although the road ahead is difficult, it is not wrong to insist on the dream of self - driving. At the same time, from the perspective of enterprise value and shareholder interests, AIGC is also a long - term and sustainable掘金赛道.