Seedance 2.5 is a video model from ByteDance, released on 31 July 2026. It generates up to 30 seconds of footage in one go, sound included - ambience, effects and score are produced in the same step as the picture. It takes up to 50 references as input: 30 images, 10 video clips and 10 audio clips, so you can feed it your own car, interior or voice. On top of that: editing a chosen region without touching the rest, and extending clips beyond 30 seconds. Below, my first attempts with this model.