Brief

Researchers have developed a new framework for diffusion synchronization, a method that coordinates multiple diffusion model trajectories to enhance generative capabilities. This approach is based on optimal control, mathematically deriving a way to guide trajectories toward coherent solutions during sampling without requiring additional training. The method has demonstrated improvements across various collaborative generation tasks and modalities, establishing a principled foundation for extending pretrained generative models. AI