Search papers, labs, and topics across Lattice.
4
0
4
Unlocking head-level control in Diffusion Transformers enables precise motion transfer without the need for retraining, revolutionizing video generation capabilities.
A 2-step generation can yield better physical consistency than a lengthy 50-step process, challenging assumptions about model depth in video synthesis.
Synthesizing realistic intermediate video frames just got a whole lot better, thanks to a novel attention mechanism that anchors to keyframes and text prompts for improved consistency and semantic alignment.
Uncover the "when and where" behind motion in video diffusion models with a new technique that visualizes motion concepts without gradients or retraining.
