ByteDance’s AI Goals Are Stalled by Computational Limits and Copyright Issues

Step aside Sora 2, a groundbreaking new AI video model has arrived.

In early February, ByteDance introduced Seedance 2.0, a significant upgrade to its previously under-the-radar video model. Its impressive capabilities took the AI landscape in China by surprise, even catching the attention of those who had been doubtful about AI-generated video, often dismissing it as low-quality content.

Feng Ji, the founder of Game Science, known for creating China’s globally acclaimed video game Black Myth: Wukong, expressed online how “deeply shocked” he was by the model’s capabilities and suggested that Seedance 2.0 could challenge China’s existing copyright laws and content moderation frameworks. Pan Tianhong, who manages a popular Chinese video production studio with over 15 million social media followers, shared a video praising Seedance 2.0 as far superior to prior video-making models. “It thinks like a director,” he remarked.

However, access to the model is currently limited. As of this week, ByteDance is allowing only existing users of its AI apps in China—including the popular chatbot app Doubao, alongside a mix of lesser-known platforms like Jimeng, Xiaoyunque, and Spark—to utilize Seedance 2.0. These applications are restricted to the Chinese market, preventing international users from exploring the model themselves. As a result, some resourceful individuals in China have begun reselling their ByteDance accounts to eager AI enthusiasts abroad.

Hints indicate that broader accessibility might be on the way. ByteDance recently updated its API platform and outlined the pricing strategy for Seedance 2.0: a 15-second video, which is the maximum length currently possible, is estimated to cost just over $2. Though ByteDance has yet to open API access to third-party developers, this may soon change.

Afra Wang, the author of the Substack newsletter Concurrent and a keen observer of the US-China AI scene, highlights that Seedance 2.0 illustrates the diverging paths taken by the two nations. Prior to the release of Seedance 2.0, several established video-making AI tools globally, like Kling AI, were developed by Chinese companies. “China hasn’t created a competent AI coding tool, which is why many Chinese users depend on tools like Claude Code or Codex; however, in terms of video AI, China is far ahead of the US,” Wang states.

Despite all the excitement, Seedance is encountering two significant challenges. Weeks after its launch, ByteDance is dealing with a compute bottleneck, resulting in hours of wait for a single video to be generated. Additionally, major film studios, such as Disney, Netflix, and Paramount, have issued cease-and-desist letters to ByteDance, claiming that the outputs from Seedance 2.0 violate their copyrights. ByteDance has not responded immediately to requests for comment.

The Bandwidth Issue

Even if you manage to access a ByteDance AI app, generating a video with Seedance 2.0 remains a challenge, as many users are attempting to do the same, and ByteDance has not yet allocated sufficient compute resources to meet the demand.

When I attempted to create a clip with one of ByteDance’s apps this week, it informed me that I was number 90,985 in the queue, with an estimated wait of about four hours for a five-second video. After two hours, the app notified me that I had six more to wait. At that point, I decided it was best to call it a night.