ByteDance AI Sparks Unauthorized Data Training Controversy in China

"Create a video of these characters fighting next to Shanghai's Oriental Pearl Tower. Reference the movie 'Transformers' style and emphasize dynamic energy."

When two photos—one of Chinese coffee chain Mixue's mascot and another of a robot wearing a Starbucks apron—were uploaded with this simple prompt, something remarkable happened. Within three minutes, a 20-second high-resolution animation emerged showing the Oriental Pearl Tower collapsing while a snowman character releases energy barriers. The blockbuster-quality production required no scene transitions or detailed instructions.

SeaDance 2.0 Debuts with 'Terrifying' Capabilities

On January 9, prominent Chinese tech influencer Tim (Fan Tianhong) released a review of ByteDance's new video AI model "SeaDance 2.0," declaring it would "upend the industry." The model, launched eight months after version 1.0, features "narrative audio" that generates natural speech synchronized to lip movements and offers 30% faster production speeds.

However, Tim used the word "terrifying" six times in his review. The reason: evidence emerged that the AI had trained on personal data without user consent.

When Tim uploaded his own photo with a simple prompt, the AI replicated his actual voice without any audio data input. It even accurately rendered the back of his company building—a view not shown in the uploaded photo.

Unauthorized Training: No Legal Recourse?

Industry observers believe ByteDance likely trained the model using vast user data from Douyin, China's version of TikTok. Tim, an influencer with over 10 million followers who has posted 428 videos on Douyin, stated: "This shows extensive learning based on my videos." He emphasized he received no notification regarding copyright use.

Current Douyin terms of service permit the company to use uploaded content globally for free for research, development, and marketing purposes, leaving users with limited legal recourse.

China Leverages Copyright Gray Zones

China is rapidly closing the gap with the United States in video AI by exploiting copyright gray areas. Kuaishou, parent company of video AI leader Kling, similarly includes provisions for AI training in its user agreements. Chinese authorities have implemented guidelines opening key public data for AI training to accelerate technological dominance.

Beijing's Internet Court has ruled that AI-generated content can receive copyright protection if users demonstrate sufficient creative effort, providing companies with grounds for aggressive development.

This stands in stark contrast to the United States and Europe, where AI copyright lawsuits continue mounting. OpenAI and Google face criticism over unauthorized training and have established opt-out guidelines for rights holders while facing regulatory investigations.

The Wilson Center, a Washington D.C.-based think tank, has warned that while the U.S. focuses on regulation, China is building asymmetric advantages through unrestricted data access.