In late July 2026, ByteDance announced Seedance 2.5, a new version of its video generation model capable of producing a 30-second clip in a single pass without stitching together separate segments. Unveiled during the keynote of the Volcano Engine FORCE conference, the model is rolling out first in China on the Jimeng and Doubao apps, with an API expected shortly on the Ark platform. The main promise is consistency, keeping the same face and style from the first frame to the last. Seedance 2.5 accepts up to 50 reference materials in a single input, including up to 30 images, 10 videos, and 10 audio clips. ByteDance claims roughly a 20% improvement in prompt adherence over the previous version. The model also supports multi-turn extension, allowing users to lengthen a sequence beyond 30 seconds by generating additional segments consistent with those already produced.