DeepSeek Moment? ByteDance’s Seedance 2.0 Challenges Sora in AI Video Generation

ByteDance’s AI video generation model Seedance 2.0 has entered limited beta testing, supporting text, image, video, and audio as reference inputs.

Users can provide up to 9 images, 3 video clips, and 3 audio files simultaneously, specifying actions, camera movements, characters, scenes, visual effects, and sound effects. Based on these references and prompts, Seedance 2.0 can synthesize videos. For privacy reasons, it currently does not support uploading images containing realistic human faces.

Seedance 2.0 can accurately replicate the composition and character details of reference images, as well as reproduce the cinematic language, complex motion rhythms, and creative visual effects of reference videos. The generated videos support smooth extensions and seamless transitions, enabling the creation of continuous shots based on user instructions. It also allows users to replace, remove, or add characters within existing videos.

Even without reference video material, a simple text prompt is sufficient. The model can automatically plan storyboards and camera movements based on the described scenario, adding effects such as slow motion or camera shake when required by the narrative. At the same time as video generation, Seedance 2.0 can produce matching sound effects and background music, with support for lip-syncing and emotional alignment.

Compared with other AI video models, ByteDance claims that Seedance 2.0 generates 2K video 30% faster than competitors such as Kling. While Sora emphasizes physical realism and Kling focuses on motion control, Seedance 2.0 specializes in producing coherent, multi-scene sequences with native audio. Following DeepSeek’s challenge to GPT, China’s large models are now mounting another direct challenge to Sora.

On social platform X, discussion around Seedance 2.0 has remained highly active.

One user, who has studied digital filmmaking for seven years, said that Seedance 2.0 is the only model that truly frightened him, claiming that nearly all roles in the film industry could disappear. According to him, 90% of the skills he learned can already be performed by Seedance 2.0.

Product designer and UX/UI designer Linus Ekenstam commented after viewing Seedance 2.0 examples: “It will break the internet. One hundred percent.”

After testing Seedance 2.0, Yocar, producer of Black Myth: Wukong, posted on Weibo praising it as “the strongest video generation model on the planet today.” He also expressed pride that Seedance 2.0 comes from China, while warning of the risks of fake video proliferation and a looming trust crisis, noting that “hyper-realistic video will soon have no barriers to entry.”

Seedance 2.0 is not without flaws. Some users reported that, even with identical prompts and reference materials, the model still struggles with fine-grained control. Issues such as garbled Chinese subtitles have also been observed.

While Seedance 2.0 may not yet fully transform the film and television industry, its emergence undoubtedly lowers the threshold for professional expression, giving more people the tools to visualize their ideas.