Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
Affiliation:
4
0
5
An 8x token reduction and 11.1x latency improvement in video generation is achieved by training autoencoders that are generation-aware, not just reconstruction-aware.
Lip Forcing achieves real-time lip synchronization at unprecedented speeds, outperforming traditional models by over 39 times while maintaining high fidelity.
Denoising directly in a 3D reconstructor's feature space unlocks simultaneous recovery of scene geometry and high-quality imagery from degraded multi-view inputs.
Get persistent world memory in your video diffusion model for free: WorldKV matches full KV-cache fidelity at 2x throughput without any training.