ByteDance launches Seedance 2.5 for 30-second video generation

ByteDance launched Seedance 2.5, a video generation model that creates 30-second clips in one go and supports multi-round extension. It builds on the unified multimodal audio-video architecture, focusing on long narratives, multimodal references, and editing. It targets film, ads, education, manufacturing, embodied AI, and autonomous driving.

The model accepts up to 30 images, 10 videos, and 10 audio clips as references, understanding composition, style, characters, and more. It keeps characters consistent even in group scenes. Examples show a singer's stage performance and a subway chase, emphasizing complete story arcs.

Seedance 2.5 will roll out on Jimeng AI and Doubao Pro, with API access via Volcano Engine. ByteDance aims to shift video generation from "making a clip" to "finishing a creation."