ByteDance Launches Seedance 2.5 Video Generation Model in China
ByteDance has introduced its Seedance 2.5 AI video generation model in China, capable of producing up to 30 seconds of video content per generation.
Chinese technology firm ByteDance has launched its latest AI video generation model, Seedance 2.5, for the domestic market. A key improvement in this version is its ability to generate video clips up to 30 seconds in length in a single run, doubling the previous 15-second capacity. This upgrade aims to facilitate the creation of longer and more complex narrative content.
Building on the architecture of Seedance 2.0, the new model emphasizes enhanced capabilities in long-form narrative generation, multimodal reference input, and editing functions. ByteDance positions Seedance 2.5 for applications across industries such as film, advertising, education, industrial manufacturing, and embodied intelligence.
The model is designed to arrange multiple logically connected shots within the 30-second output, enabling content that follows a narrative structure. Seedance 2.5 also supports iterative video extension, allowing users to build upon existing generated content while maintaining consistency in characters, scenes, and audio elements.
Seedance 2.5 can process a variety of reference materials, including up to 30 images, 10 video clips, and 10 audio segments simultaneously. The AI is engineered to interpret elements from these diverse inputs—such as composition, style, and characters—and integrate them into the video generation process according to user prompts. The model is also capable of rendering scenes with multiple individuals and maintaining the stability of character likenesses.