VideoCoCo introduces an agentic dual-engine framework for physically consistent text-to-video generation. A coding agent converts a prompt into executable Blender code that specifies both the scene and its temporal evolution. Blender then produces a deterministic spatiotemporal draft, which a generative video engine transforms into a photorealistic result through draft-conditioned editing. The authors also build VideoCoCo-3K, a dataset of draft-instruction-target triplets for adapting the editor to simulated inputs. Against the OmniWeaving baseline, VideoCoCo improves PhyGenBench from 0.475 to 0.558 and VBench-2.0 from 52.18 to 77.88.
No heat snapshots are available in the last 24 hours.