ByteDance’s Seedance 2.5 targets editable 30-second AI video

ByteDance launched Seedance 2.5 on 31 July with up to 30 seconds of synchronised audio-video in one generation. The release emphasises reference-based creation and more precise editing, including timestamp-level instructions, green-screen replacement and camera-perspective changes.

The model can use up to 30 image references, alongside video and audio references. That is closer to a creative brief than a single prompt: teams can specify the look, characters, pacing and selected changes without recreating every shot from scratch.

Timestamp control should be treated as directional rather than frame-perfect. Build a production workflow around clear shot ranges, approval points and conventional editing tools, then evaluate consistency and rights risk before comparing models on headline duration alone.