Search papers, labs, and topics across Lattice.
1
0
3
2
Surprisingly, high sparsity in video diffusion models doesn't degrade generation quality if the sparse mask accurately mimics the tile-wise geometry of full attention.
