Seedance 2.0 frames video generation as a world-complexity problem, not a rendering one
A native multimodal audio-video generation model whose stated target is complexity in the depicted world rather than fidelity in the depicted frame. That reframing is the substance of the paper.
Most video generation progress has been measured on appearance — resolution, temporal coherence, artefact rate. Framing the problem as world complexity moves the target to whether the depicted scene behaves consistently: objects persisting, causes preceding effects, physics staying put across a cut.
Native audio-video is the architectural half. Generating both in one model rather than aligning separately produced streams removes the seam where lip sync and impact timing usually drift.
The evaluation problem is unsolved and the paper does not pretend otherwise. There is no accepted metric for whether a generated world is internally consistent, which means progress on the stated goal is currently assessed by watching.
arXiv — Seedance 2.0: advancing video generation for world complexity → · arXiv — Evolution of video generative foundations →