Search papers, labs, and topics across Lattice.
3
0
6
Image editing models leak fascinating hints about their world knowledge through "edit spillover"—unintended changes to semantically related regions—and this paper turns that leakage into a probe.
Achieve 16% faster DiT inference by caching the right features at the right denoising steps, without sacrificing image quality.
Stop identity features from bleeding into your backgrounds: SpatialID uses attention-based masks to inject personalized visual features only where they matter, without any training.