Search papers, labs, and topics across Lattice.
2
0
4
State-of-the-art generative models struggle to maintain physical consistency and coherent interactions over time, revealing critical gaps in their world modeling capabilities.
Long video generation fails not just because of limited context length, but because of *how* that context is allocated – and ReCA's hierarchical approach shows a way to fix it.
