Search papers, labs, and topics across Lattice.
3
0
7
State-of-the-art video generation and editing now hinges on a surprisingly simple division of labor: MLLMs for semantic planning, diffusion models for photorealistic rendering.
Forget noisy samples, RL can now directly optimize the *gradients* of diffusion distillation, leading to SOTA few-step image generation.
Autoregressive image generation gets a serious upgrade: Generative Refinement Networks use near-lossless quantization and a global refinement mechanism to beat diffusion models in image quality and efficiency.
