Today, Doubao video generation model Seedance 2.0 was officially released and fully integrated into Doubao and Jimeng products. It was also launched in the Volcano Ark Experience Center. at present,Seedance 2.0 limits real person images/videos as subject reference. If you need to use real people as subject reference, you must be verified or authorized by yourself.
According to ByteDance, in order to comprehensively evaluate the model's comprehensive capabilities in multi-modal scenarios, the team collaborated with experts in the film and television field to establish a comprehensive evaluation set and related evaluation standards covering audio and video generation, reference and editing scenarios.
Vincent video ability assessment
Tusheng video ability evaluation
Multimodal reference generation evaluation
In various evaluations, the comprehensive performance of Seedance 2.0 has reached the industry-leading level.
However, ByteDance also admitted frankly,The model still has room for optimization in terms of detail stability, multi-person demographic matching, multi-subject consistency, text restoration accuracy and complex editing effects.
Officially stated,Seedance 2.0 is far from perfect, and its generated results still have many flaws.It will continue to explore the in-depth alignment of large models and human feedback, so that more efficient, more stable, and more imaginative audio and video production tools can serve more creators.