ByteDance’s AI Push Stalled by Compute Limits and Copyright Issues
▼ Summary
– ByteDance has released Seedance 2.0, a powerful new AI video model that has shocked China’s AI ecosystem with its advanced, director-like capabilities.
– Access to Seedance 2.0 is currently restricted to users of ByteDance’s domestic Chinese apps, preventing international testing, though API pricing suggests broader access may be coming.
– Industry experts in China believe the model’s power poses significant challenges to existing copyright regulations and content moderation systems.
– An analyst notes this release exemplifies a divergence in US-China AI development, with China being far ahead specifically in video-generation AI.
– The model faces serious operational problems, including a compute bottleneck causing long wait times and cease-and-desist letters from major movie studios over copyright infringement.
A powerful new contender has emerged in the world of AI-generated video, capturing significant attention within China’s tech ecosystem. ByteDance’s Seedance 2.0 model represents a major leap forward for the company’s video generation capabilities, impressing even seasoned skeptics of the technology. Its sophisticated output has prompted notable figures in China’s creative industries to voice both admiration and concern. Feng Ji, founder of the studio behind the hit game Black Myth: Wukong, described being “deeply shocked” by the model’s potential, suggesting it could challenge existing copyright and content moderation frameworks. Similarly, prominent video producer Pan Tianhong praised the model’s directorial intelligence, stating it surpasses all previous video-making tools.
Despite the excitement, widespread access to Seedance 2.0 remains tightly controlled. Currently, only users of ByteDance’s domestic Chinese AI apps, such as the popular Doubao chatbot and other platforms like Jimeng and Xiaoyunque, can experiment with the model. This geographical restriction has created a curious gray market, with some users inside China reselling their account access to eager international adopters. However, a broader rollout appears imminent. ByteDance recently updated its developer platform with proposed pricing, indicating that generating a 15-second video, the current maximum length, would cost just over two dollars. While API access for third-party developers is not yet available, it is expected to open soon.
The development of Seedance 2.0 highlights a fascinating divergence in the global AI race. Observers like Afra Wang, who authors the Concurrent Substack on U.S.-China tech, note that China has established a clear lead in AI video generation, with tools like Kling AI also originating there. This stands in contrast to other AI domains, such as coding assistants, where Chinese developers still largely rely on Western tools like Claude Code. “When it comes to video AI, China is miles ahead of the US,” Wang states, underscoring the specialized strengths emerging from different innovation environments.
Yet, behind the promising technology lie significant hurdles. ByteDance is grappling with a severe computational bottleneck, struggling to meet the overwhelming demand from users. Attempts to generate a video currently place users in queues of tens of thousands, with wait times extending to several hours for just a few seconds of content. This infrastructure constraint severely limits practical testing and application. Concurrently, the company faces mounting legal pressure. Major Hollywood studios, including Disney, Netflix, and Paramount, have dispatched cease-and-desist letters alleging that Seedance 2.0’s outputs infringe on their copyrighted intellectual property. These dual challenges of scalable compute power and copyright compliance present substantial obstacles that ByteDance must navigate to fully realize its model’s potential and move beyond its current limited, invitation-only phase.
(Source: Wired)
