Paper summary: Seedance 2.0: Advancing Video Generation for World Complexity

Problem

Video generation with synchronised audio, guided by many kinds of references, has usually needed several separate systems.

Method

Seedance 2.0 uses a unified, large-scale architecture for joint audio–video generation. It accepts text, image, audio and video inputs and supports a wide range of reference and editing capabilities. A Fast variant targets low latency. According to the abstract, it was released in China in early February 2026.

Key results

From the abstract:

• generates audio-video clips of 4 to 15 seconds at 480p or 720p
• the open platform accepts up to 3 video clips, 9 images and 3 audio clips as references
• in expert evaluations and public user tests, performance on par with the leading level in the field

Limitations

There’s no dedicated limitations section in the extracted text. The evaluation is described in general terms, without specific benchmark numbers in the abstract.