Key facts
- Chinese AI video models lead in visual quality but lag in safety safeguards.
- ByteDance's Seedance 2.0 Pro ranked highest for visual quality.
- Google DeepMind's Veo and OpenAI's Sora also ranked among leading models.
- Training data availability and less strict copyright adherence may contribute to Chinese models' lead.
- Industry professionals praise the detail, sound synchronization, and actor stability in Chinese AI video tools.
Chinese artificial intelligence video-generation models have surpassed their US counterparts in visual quality, though they lag in safety safeguards, according to a study by researchers at the University of Hong Kong. ByteDance's Seedance 2.0 Pro achieved the highest score for visual quality among 12 leading models, outperforming Google DeepMind's Veo and OpenAI's Sora. The study, led by Professor Zhenhui Jack Jiang, suggests that Chinese models benefit from vast training data, potentially with fewer copyright restrictions, contributing to their advanced realism and intuitiveness. Industry professionals have noted the superior detail, sound synchronization, and stable virtual actors in these Eastern tools. However, concerns about intellectual property and safety controls are prompting promises of stricter oversight from developers.