ByteDance Releases Seedance 2.5 Video Generation Model

ByteDance has officially launched Seedance 2.5, marking a major update to its video generation model family. Built on the multimodal audio-video joint-generation architecture established in Seedance 2.0, the new release centers on foundational generation and reference-based creation to improve long-form storytelling and editing precision.

30-Second Single-Pass Generation and Extensions

A primary upgrade in Seedance 2.5 is the extension of single-pass video generation from 15 to 30 seconds. Within a single generation, the model can organize multiple logically connected shots to unfold a narrative featuring setup, development, turning points, and resolution. Additionally, Seedance 2.5 supports multi-round extensions, allowing creators to smoothly append subsequent shots while maintaining the consistency of characters, environments, and narrative pacing to produce multi-minute content.

Upgraded Multimodal Referencing

The model significantly expands its multimodal reference capabilities, enabling users to input up to 50 reference materials in a single pass—specifically comprising up to 30 images, 10 video clips, and 10 audio clips. This wider variety of references allows the system to better capture creative intent across complex scenarios involving multiple subjects, diverse scenes, and flexible camera movements. Furthermore, it enhances specific reference workflows including clay render referencing, motion references, and creative stylization.

Precise Editing and Professional Controls

Seedance 2.5 introduces timestamp-level control for targeted editing of audio and video content during and after generation. Creators can adjust narrative pacing, camera perspectives, and specific actions within designated time frames while preserving continuity. Advanced professional capabilities, such as green screen replacement and reference-based editing, have also been refined to support complex production environments in film and advertising.