ByteDance Previews Seedance 2.5 — Single Native 30-Second 4K Clips and Up to 50 Reference Inputs in One Generation

The Next Web·medium signal

At the Volcano Engine FORCE conference (June 23), ByteDance previewed Seedance 2.5, claiming a single native 30-second video clip (breaking the typical sub-10s barrier without stitching), up to 50 reference inputs in one generation, and region-level editing that redraws part of a frame without touching the rest. The model is currently in enterprise beta with public launch targeted for early July; Seedance 2.0 was simultaneously updated to support 4K output. If the 30-second native-generation claim holds, it pressures Runway, OpenAI's Sora, and Google Veo on long-form coherence.