1 paper · 1 filter
Kaichen Zhou, Yuzhen Chen, Fangneng Zhan +8
Video world models can generate realistic futures from a single instruction, but they often fail to track the same physical points consistently across time. As a result, the genera…