Summary
ByteDance has officially unveiled Seedance 2.5 at its Volcano Engine summer conference in Beijing, and the announcement marks a significant leap forward for AI video generation. Revealed by Volcano Engine president Tan Dai, the new model is currently in global enterprise beta and is expected to launch publicly in early July 2026.
The headline upgrade is native 30-second single-shot video generation — double the length of Seedance 2.0 and every other current competitor — without requiring stitching of separate clips. To maintain visual consistency across longer sequences, the model now supports up to 50 reference materials simultaneously (up from roughly 12–15), spanning images, video, audio, and text. A third new feature enables partial scene editing, allowing changes to one region of a frame while the rest of the scene remains stable. ByteDance also announced that the existing Seedance 2.0 is being upgraded to support 4K native output.
Beyond the video product itself, Tan Dai positioned Seedance as a stepping stone toward world models — AI systems that understand physical reality — and noted real-world deployment in embodied AI, robotics, industrial manufacturing, and self-driving car simulation. The company reports that its DAI models collectively handle over 18 trillion tokens of usage per day, a 1,500x increase since launch two years ago. The announcement also included a new AI copyright and licensing platform developed in partnership with Hong Kong filmmaker Stephen Chow.
📺 Source: TheAIGRID · Published June 23, 2026
🏷️ Format: News Analysis
