August 1 News.ByteDanceOfficial launch of a new generation of video creation models Seedance 2.5. The new model has been overhauled in four dimensions: long generation time, multimodular understanding, editorial precision and language adaptation, with the aim of enabling AI VideoThe idea is to create a “subsidiary innovation” to the “industrial productivity infrastructure”。
A single video generation time increase from 15 seconds to 30 seconds and support multi-wheel extensions of up to 90 seconds
A single user can enter up to 30 pictures, 10 videos and 10 audio as reference material
A white model reference and a green curtain reference feature have been added。
In terms of editorial precision, Feedance 2.5 introduces a time-stamp-based video local editing capability that supports targeted changes and consistency in the role, action or sound of the given clip. In addition, the model supports more than 10 languages and enhances compliance with complex directives。
At present, Seedance 2.5 is on line, i.e. dream, button and little sparrows, to jump under the flag and will open the API service to business users via the volcanic engine. Byte beats indicate that longer coherent narratives and greater physical world understanding will continue to be explored in the future。