Search papers, labs, and topics across Lattice.
This paper introduces latent spatial memory for video world models, which allows for the storage of scene information directly in the diffusion latent space, circumventing the need for computationally expensive pixel-space reconstructions. By employing a novel framework called Mirage, the authors achieve significant improvements in both speed and memory efficiency, with up to 10.57 times faster video generation and a 55 times reduction in memory footprint compared to traditional methods. The approach leverages depth-guided back-projection and latent-space warping, resulting in state-of-the-art performance on benchmark tasks while preserving rich features of the learned representations.
Latent spatial memory can accelerate video generation by over 10 times while dramatically reducing memory usage, revolutionizing how we model dynamic scenes.
Video world models that maintain 3D spatial consistency across generated frames typically rely on explicit point cloud memory constructed in RGB space. This design is both computationally expensive, requiring repeated rendering and VAE encoding, and inherently lossy, as the round trip through pixel space discards rich features of the learned latent representation. In this paper, we introduce latent spatial memory for video world models, a persistent 3D cache that stores scene information directly in the diffusion latent space, avoiding pixel-space reconstruction. Building on this, we propose Mirage, a latent-space spatial memory framework that constructs the memory by lifting latent tokens into 3D via depth-guided back-projection and queries it by synthesizing novel views through direct latent-space warping. This unified formulation eliminates both the information loss of pixel-space reconstruction and the computational burden of repeated encoding and rendering. Experiments show that latent spatial memory achieves up to 10.57times faster end-to-end video generation and 55times reduction in memory footprint relative to explicit 3D baselines. Leveraging the geometric prior of the diffusion model, Mirage attains state-of-the-art performance on WorldScore and strong reconstruction quality on RealEstate10K.
Microsoft Research