Written by | Hao Xin

Edited by | Wu Xianzhi

When an AI company grapples with a multitude of challenges—ranging from internal management upheaval and technological roadblocks to profitability pressures and the imperative to go public—how can it break free from the shackles?

OpenAI's response comes in the form of Sora2, a solution that redirects the disappointment stemming from the 'incremental improvements' of ChatGPT-5 towards envisioning a future where AI reshapes a Douyin-like platform.

Unlike the previous endeavors by tech giants like Google and Kuaishou, which primarily focused on enhancing model performance and AI video generation capabilities, Sora2 stands out as the inaugural platform that seamlessly blends the attributes of content creation tools with those of short video consumption. It diverges from the path taken by AI video generation tools such as Kling and Jime, which initially launched as web-based services before transitioning to apps. Instead, Sora2 made its debut as an independent application on the Apple App Store.

Although the 'invitation-only' marketing tactic may seem like a relic of the past, it has effectively sparked a buzz around Sora2. According to Appfigures data, the Sora app garnered 627,000 iOS downloads in the United States and Canada during its inaugural week through an invitation-based system, surpassing ChatGPT's initial launch performance and swiftly climbing to the top of the free app chart on the U.S. Apple Store.

Recently, OpenAI appears to have significantly eased its restrictions on invitation codes. Multiple users have reported that while an invitation code was previously mandatory for logging in and using the app, they can now access it without one after waiting a few days and reopening the app.

Our tests corroborate this observation—upon initial login, the interface still prompts for an invitation code. However, after waiting approximately 20 minutes and attempting again, the system directly navigates to the app's main page without any verification.

This shift mirrors a recent decline in user enthusiasm. Currently, Sora2 resembles Douyin's nascent stages as A.me or TikTok's Musical.ly phase, characterized by ambiguous positioning, high substitutability, and a lack of sustained allure.

Sora2 now stands at a pivotal juncture—it can either surmount its early limitations and emerge as a true industry leader, akin to Douyin, or succumb to acquisition and merger, gradually fading from the market as a transitional product in the annals of AI video history.

New Possibilities

The Sora 2 app description reads, 'Transform your thoughts into videos and immerse yourself in the experience,' a sentiment that encapsulates its core offering.

Here, 'you' is intentionally left vague, distinguishing it from the previous phase of AI video tools tailored for professional creators. OpenAI clearly aims to capture a broader consumer market, aspiring to create another 'blockbuster' AI application like ChatGPT.

The design philosophy of Sora 2 echoes that of TikTok, attempting to coax content consumers into becoming content producers. Lowering the barrier to entry becomes paramount, eliminating the cumbersome operations of AI video tools and retaining only the function of generating videos with a single sentence. Additionally, pre-set animated and gaming characters replace the need for actual face and voice authorization, simplifying content creation.

The tool-based nature of AI video inherently limits its dissemination attributes. For instance, the widely popular 'Venom' special effects, which went viral on Douyin, are rarely known to have originated from Pixverse. This results in insufficient user loyalty to content creation tools, as users flit between several similar tools without lasting engagement, essentially squandering resources.

In the Sora2 app, users can also like, follow, and comment on videos—though all content is AI-generated. Algorithmically, Sora2's recommendation system can be fine-tuned through natural language instructions and incorporates user activity within the app, location information, post interactions, and ChatGPT conversation history under the same account into its personalized recommendation algorithm to enhance interactivity.

Sora2 already exhibits some characteristics of a 'lively' platform, with genuine and continuously increasing likes and comments. A video lasting only five seconds may receive tens of thousands of likes and hundreds of comments. The viral nature of content is also spreading outward, with videos labeled 'Generated by Sora2' going viral on platforms like YouTube, reaching 7.9 million views. Compared to AI video tools that have fizzled out or remained niche, this figure is impressive, demonstrating the future potential of AI video.

If Douyin's 'Record Beautiful Life' encourages users to become content consumers and sharers, then Sora 2's Remix function propels AI video creation into a new phase where 'creators and consumers jointly shape content.'

In traditional video creation, creators produce content, and users consume it passively, with interaction limited to likes and comments. Sora 2's Remix function completely breaks this mold—users are no longer mere spectators but can directly engage in secondary, tertiary, or even multiple iterations of creation based on the original video, allowing content to evolve continuously during dissemination.

In a video described as 'My grandmother's real footsteps,' depicting an elderly woman driving away crocodiles step by step, swiping left reveals users replacing the crocodiles with hippos, giraffes, dinosaurs, tanks, and other elements. Data shows that the total number of secondary creations based on the original video reached 18,000, with tertiary creations involving hippos and giraffes numbering between 6,000 and 8,000.

This means a simple idea can, with AI assistance, spawn thousands of variations, exponentially increasing the richness of the content ecosystem and deepening user engagement with the AI application.

The Competitive Phase with 'TikTok'

Sora2's ambition to emulate Douyin is evident, but it is still in a very early stage.

In 2016, Zhang Yiming identified short video as the next big trend in content entrepreneurship and decided to invest heavily in this area. That same year, Douyin's predecessor, A.me, was launched and later renamed 'Douyin,' focusing on a music video community for young people.

Through comprehensive promotional strategies, including celebrity endorsements, program sponsorships, and Spring Festival Gala collaborations, Douyin rapidly increased its brand awareness and market share. Particularly, collaborations with celebrities and high-traffic artists attracted a large number of new users in its early stages.

In 2017, ByteDance acquired the music video platform Musical.ly, laying the foundation for the international version, TikTok.

Today, Sora2 is in a phase between exploration and promotion, requiring technological validation and scenario anchoring. As an emerging technology, AI video will face competition with TikTok for a considerable time.

From the perspectives of ordinary users and professional creators, Sora2 currently feels like a work in progress, with issues such as unstable traffic, inconsistent generation results, and poor user experience.

Some users told us that while one of their Sora2 videos received over 100,000 views, another video created with Sora2 garnered 8.7 million views on TikTok. In comparison, they still find TikTok more attractive.

For professional creators, Sora2's appeal lies in its lack of generation limits, allowing them to quickly present a short film prototype integrating sound effects, visuals, and narration. However, the drawbacks are also evident, with too many restrictions, such as 'no content with copyright information, mandatory watermarks, and increasingly strict template limitations.' Multiple attempts are often needed to achieve the desired effect, and the generation quality falls short of other contemporary AI video tools.

The competition between TikTok and AI video tools is not just about users but also about commercial markets.

At its core, this competition revolves around the complete chain of 'user attention-content productivity-commercial monetization,' with both platforms vying for market dominance across multiple dimensions and levels. User choices reflect not only personal preferences but also votes for optimizing the commercial ecosystem. Ultimately, the platforms' competition for users points to control over the commercial market.

AI video tools, represented by Sora 2, aim to shape an AIGC+UGC ecosystem, lowering creation barriers through AI technology and enabling users to produce high-quality videos. The Remix function allows users to engage in secondary creation based on existing templates, inheriting the original video's style and structure while generating differentiated content through key element modifications. IP collaborations further enrich content appeal by introducing well-known characters.

In terms of business models, there is also potential overlap and competition between the two sides in the future.

When AI video companies incorporate 'social interaction' into the creation process, their traffic acquisition logic intersects with Douyin's UGC model. When TikTok uses AI tools to lower creation barriers for ordinary users, its service boundaries inevitably extend into the 'professional content production' domain where AI video companies excel.

In e-commerce marketing, both sides aim to enhance product conversion rates through high-quality video content. Whether it's Sora 2's AI-generated marketing videos or Douyin's live-streaming e-commerce, both essentially compete for merchant budgets and user attention.

Who Will Be China's Version of Sora2?

It remains uncertain whether we need an independent AI version of 'TikTok' or an advanced AI version of 'TikTok' in the future.

Sora2's aspiration to replicate TikTok's success is not merely a technical challenge but a comprehensive consideration involving copyright, IP, computing power, traffic, and other factors.

Copyright issues are the primary obstacle on Sora 2's development path. In video creation, the copyright ownership of materials is complex and diverse. Sora 2 needs to invoke (use) a large number of images, audio, and video clips as basic elements when generating videos. However, most of these materials are strictly protected by copyright, and obtaining legal authorization requires not only paying high fees but also navigating cumbersome legal procedures and complex negotiations. Previously, OpenAI and China's MiniMax have already faced lawsuits over copyright issues.

Introducing an IP revenue-sharing mechanism that allows users to legally use well-known IP characters to generate videos can stimulate user creativity and bring more diverse content to the platform. However, this also faces strict requirements from IP copyright holders and complex collaboration models. Copyright holders have stringent regulations on the scope, methods, and duration of IP usage.

Traffic is a crucial factor for the survival and development of internet platforms. To achieve success, Sora 2 must attract a large number of users and maintain their activity and loyalty. However, the current short video and video creation landscape has already formed a fiercely competitive traffic environment. Head platforms like TikTok have accumulated massive user bases and strong brand influence, occupying most of the market share.

Based on the aforementioned criteria, the most competitive companies in China are still ByteDance and Tencent.

Douyin and TikTok are already two globally successful short video platforms, with core competencies in algorithmic recommendations, user growth, and commercial monetization. ByteDance is more likely to create an 'AI-advanced version of Douyin,' where AI becomes a super tool for Douyin creation rather than disrupting the existing model. Its existing editing tool, 'CapCut,' and AI video tool, 'Jime,' can be integrated with Douyin, breaking down barriers between professionals and non-professionals and allowing users to easily generate AI videos 'like using filters' without needing to download a new app.

Douyin's 'moat' is difficult to challenge, as users are already accustomed to the 'video browsing + creation' mode. AI video generation can seamlessly integrate into the existing ecosystem, such as allowing users to input text descriptions and automatically generate short videos.

In terms of copyright, Douyin has already established copyright libraries for music, film, television, and materials, collaborating with major record labels and film companies to ensure users can legally use music and video clips during creation. For example, Douyin's 'CapCut' tool already provides copyrighted music, filters, and effects that users can directly access.

Tencent's strengths lie in social networking (WeChat, QQ), IP (games, animation), and financial payments (WeChat Pay). If Tencent were to venture into AI video generation, it would more likely create an independent AI video platform rather than directly modifying existing products.

Tencent owns top-tier IPs such as 'Honor of Kings,' 'Peacekeeper Elite,' 'Fox Spirit Matchmaker,' and 'Soul Land.' These IPs can be directly licensed to AI video generation platforms, allowing users to create 'AI Honor of Kings short dramas' or 'AI Soul Land animations.' In terms of film, television, and music, Tencent Video, China Literature, and QQ Music provide rich copyrighted material libraries for AI video generation platforms, even exploring AI-generated adaptations of online novels into short dramas.

Additionally, Tencent excels in the 'social + content' model, such as allowing users to generate AI videos and share them instantly in WeChat groups or Moments, even combining WeChat Pay for 'AI video-driven e-commerce.'

'Technology and gameplay are easily replicable; the future of AI video development will still be controlled by major players.'

The 'bottleneck' issues that truly determine the direction of the AI video industry and become key constraints for enterprise development lie in the underlying resources and ecological barriers held by major players. These are the 'invisible moats' that AI video companies find difficult to breach.