Seedance 2.0 Photo: VCG

Chinese tech giant ByteDance officially launched its artificial intelligence (AI) video-generation model Seedance 2.0 on Thursday, while imposing strict restrictions on uploads featuring real-person images and videos, according to a statement the company sent to the Global Times.

The release followed several days of beta testing of the model, sparking widespread discussion over its powerful video-generation capabilities. An X user named "levelsio" posted a short film generated by Seedance 2.0 on Thursday, comparing it with clips produced by Google's Genie 3. The visibly more realistic and richly detailed visuals drew attention online, with Tesla CEO Elon Musk commenting, "This is happening too fast."

Liu Gang, chief economist of the Chinese Institute of New Generation Artificial Intelligence Development Strategies, said the gap between Chinese and leading overseas video-generation models is rapidly closing, with some capabilities already pulling ahead.

Feng Ji, CEO of Game Science and producer of the hit game Black Myth: Wukong, described Seedance 2.0 as "the strongest video-generation model on the planet" after testing it. He noted in his Sina Weibo post that the model represents a leap forward in multimodal understanding and integration.

Liu said that the breakthrough is no coincidence, but the result of long-term data accumulation and technological evolution, adding that massive short-video datasets and continuous iteration in real-world application scenarios have strengthened its engineering foundation. 

"At the same time, China's systemic innovation in large-language model training paradigms and structural optimization has accelerated overall progress, enabling domestic players to approach — and in some aspects exceed — global competitors," he noted.

The Seedance 2.0 team said that the model is still not perfect and its outputs have flaws, adding that they will keep improving it based on user feedbacks to provide more stable and creative tools for content creators.

Seedance 2.0 is now available on Doubao's mobile app, desktop client and Web version, as well as on the Dreamina app and web platform. However, its powerful video-generation capabilities have raised concerns. In his post, Feng warned that highly realistic fake videos could soon be generated with almost zero barriers, posing challenges to existing intellectual property and content review systems.

In response, ByteDance said that on the Doubao and Dreamina mobile apps, users who wish to generate AI videos featuring themselves must first complete identity verification through audio and video recordings to create a digital avatar. On desktop and Web versions of Doubao and Dreamina, the platforms clearly state that uploading real human facial materials is currently not supported.

Chinese authorities have issued multiple regulations to prevent risks related to personal privacy and intellectual property infringement, including requiring explicit and implicit labeling of all AI-generated content and prohibiting the generation of illegal or harmful information.

Feng closed his post by expressing relief that Seedance 2.0 was made in China. Wang Peng, an associate research fellow at the Beijing Academy of Social Sciences, said that Feng's sentiment reflects a strategic confidence in China's technological governance.

"China's AIGC registration and labeling system provides legal boundaries for such a powerful tool [generative AI], giving Chinese AI management standards greater influence on the global stage," Wang noted.