ByteDance showcased its latest video generation model, Seedance 2.5, at the Volcano Engine FORCE conference on June 23rd. The model is expected to officially launch in early July, and its headline feature is a doubling of native video output length to 30 seconds. It also supports importing up to 50 full-modal reference materials at once, addressing the common pain point of character and product detail distortion in AI-generated video. Seedance 2.5 additionally allows for localized fine-tuning — editing specific details without affecting the rest of the scene.
In Seedance 2.5, Volcano Engine collaborated with professional directors to create what it calls a "3D white model preview" feature — a first in the industry. This allows creators to perform rough 3D edits and visual previews before generating the final video, aiming to improve the industrial delivery efficiency of film and TV production.
For the same event, Seedance 2.0 was also upgraded to support native 4K video generation.