Search papers, labs, and topics across Lattice.
1
0
3
Image generation models can now achieve state-of-the-art fidelity with up to 64x fewer tokens, thanks to a novel masking strategy that prevents latent space collapse.
