Seedance 2.0 Pro by ByteDance Outperforms Veo and Sora in AI Video Quality

The Seedance 2.0 Pro model from ByteDance took first place in video generation quality among 12 models in a study by the University of Hong Kong, scoring 3.84 out of 5. It outperformed Veo 3.1 from Google DeepMind with a score of 3.47 and Sora 2 Pro from OpenAI, which received 3.27. Seedance also led in visual consistency, physical realism, aesthetics, and accuracy of instruction execution. When evaluating video along with sound, it retained the top spot with a score of 3.86 compared to 3.51 for Veo and 3.35 for Sora.
High generation quality does not guarantee effective protection against harmful content. In separate safety testing, Seedance ranked sixth among 11 models with a score of 3.12. The first place went to Hailuo 2.3 from MiniMax with 3.63 points, while Veo 3.1 took the second spot with 3.62. Researchers tested how models respond to requests related to misinformation, illegal content, violence, discrimination, and copyright infringement.
The authors of the study also found that most models perform worse in generation when a reference image is added to the text request: the result of the Chinese Wan 2.6 dropped from 3.41 to 2.75 points. According to the researchers, further development of AI video requires not only improving realism and accuracy of request execution but also strengthening safety mechanisms.
Earlier, media reported that OpenAI might transfer Sora to an unnamed Hollywood startup.
This material is prepared solely for informational purposes and does not constitute financial advice or a recommendation.





