[ByteDance Follows MiniMax H3 with the Release of Seedance 2.5: Generates 30 Seconds in One Go, Supports Up to 50 Reference Materials] According to monitoring by Beating Insights, ByteDance's Seed team has launched the video model Seedance 2.5. It continues the architecture of combining text, images, videos, and audio as inputs, extending the single-generation duration from 15 seconds to 30 seconds, with support for further video extensions. The model can arrange multiple shots and complete storylines within 30 seconds. Users can iteratively extend content based on existing results while maintaining consistency in characters, scenes, audio, and narrative pacing, ultimately stitching together videos several minutes long. Seedance 2.5 allows for the input of up to 30 images, 10 video clips, and 10 audio clips at once. The new model also supports timestamp-based control and video editing. Users can specify actions or shot transitions at specific seconds, or adjust only characters, audio, or camera movements without redoing entire segments. Seedance 2.5 is gradually being rolled out on JiMeng AI and Doubao Pro, with API integration into Volcano Ark expected soon. [Original Link]