Search papers, labs, and topics across Lattice.
3
5
5
4
Today's visual generation models are often evaluated on the wrong things, leading to inflated performance claims that mask critical failures in spatial reasoning, temporal consistency, and causal understanding.
Time conditioning, a seemingly crucial component of diffusion models like DDIM, can be entirely bypassed without sacrificing generation quality by carefully shaping the evolution of noisy data manifolds.
By decoupling multimodal reasoning from high-fidelity synthesis, Query-Kontext achieves strong image generation and editing results, even outperforming task-specific SOTA methods in some cases.
