ByteDance's Seedance 2.5 Doubles Video Length to 30 Seconds, Adds 50-Reference Input and Timestamp-Level Editing
- ByteDance's Seedance 2.5 generates synchronized audio and video up to 30 seconds in a single pass, double the 15-second ceiling of Seedance 2.0.
- The model accepts up to 50 references per clip (images, video, audio, text combined), up from the prior limit of 3 videos, 9 images, and 3 audio clips (15 non-text items); ByteDance has not disclosed per-type sub-limits and text also counts toward the 50.
- Dreamina will add an editing feature letting users select a specific timestamp or person within a generated video and revise just that portion instead of regenerating the whole clip.
- Seed's model page states a generated video can be extended twice, but does not specify how many seconds each extension adds, so total achievable length is unclear.
- ByteDance claims improvements in scene transitions, character consistency, motion stability, and visual realism, but the announcement provides no benchmark figures or comparison data to support these claims.