On July 31, ByteDance unveiled its cutting-edge video generation model, Seedance 2.5, and promptly introduced it across platforms like Doubao and Jimeng. This model has witnessed extensive enhancements in core aspects, encompassing generation length, multimodal comprehension, and editing accuracy. The duration of a single video generation has been elongated to 30 seconds, accommodating joint input of up to 50 multimodal resources. Additionally, it has incorporated the functionality for localized editing, allowing for precise adjustments to specific elements without disrupting the overall visual flow.

Seedance 2.5 inherently supports over 10 languages, catering to the needs of multilingual industrial production. It will soon extend API services to enterprise users via Volcano Engine. Several enterprises, including XCMG Group, XPENG Motors, Zhiyuan Robotics, Lingchu Intelligence, Weifen Zhifei, Qiongche Intelligence, and XsparkAI, have expressed cooperation intentions and are set to be among the initial adopters of this model.

Presently, Seedance 2.5 has been deployed across diverse industrial scenarios, such as industrial operation training, intelligent driving training data generation, and embodied intelligent robot training. This marks a significant shift in video generation technology, transforming it from a mere creative tool into a powerful productivity enhancer.