Motion Dreamer: Realizing Physically Coherent Video Generation through Scene-Aware Motion Reasoning

Tianshuo Xu^1,*, Zhifei Chen^1,*, Leyi Wu¹, Hao Lu¹, Yuying Chen², Lihui Jiang², Bingbing Liu², Yingcong Chen^1,3,†

¹HKUST(GZ), ²Noah's Ark Lab, ³HKUST

* Equal Contribution, † Corresponding Author

TL;DR

We introduce Motion Dreamer, a two-stage video generation framework that decouples motion reasoning from high-fidelity video synthesis, addressing the challenge of producing physically coherent videos.

Overview

TODO

Tasks	Status
Release the collected driving data	[ ]
Release training/evaluation code	[ ]
Release the official model	[ ]