Overview
- ByteDance announced the domestic launch of Seedance 2.5, which doubles single-generation length from 15 seconds to 30 seconds and supports multi-round continuation to create contiguous multi-shot sequences.
- Seedance 2.5 accepts up to 30 images, 10 videos and 10 audio clips per request and can preserve multi-person appearance and voice consistency across generated segments.
- Volcano Engine will make Seedance 2.5 available to business users via near-term APIs, and several firms including XCMG, XPeng and multiple robotics and AI vendors have reported plans to pilot or integrate the model.
- ByteDance positions the model to cut the time and cost of producing instructional and creative footage and to generate training and simulation data for tasks like robot motion prediction and autonomous-driving corner-case scenarios.
- Seedance 2.5 builds on the Seedance 2.0 unified audio-video architecture and reflects a wider shift in generative AI toward longer, controllable multi-modal media; announced partner work is framed as early integration plans rather than completed large-scale deployments.