Search papers, labs, and topics across Lattice.
4
0
7
15
Real-time, high-resolution video editing is now possible on a single consumer GPU, thanks to a novel hybrid diffusion transformer and system-level optimizations that achieve 24 FPS at 1280x704.
By structuring diffusion-based driving models around a "scaffold" of frozen structural tokens, Fast-dDrive achieves a 12x speedup over autoregressive baselines while improving trajectory accuracy.
Scaling diffusion model alignment just got a whole lot cheaper: Sol-RL uses FP4 rollouts to accelerate training convergence by up to 4.64x without sacrificing performance.
Swap out slow, one-token-at-a-time generation in VLMs for a 6x speed boost, without sacrificing quality, using a surprisingly simple direct conversion to block-diffusion decoding.
