Abstractaligment problem?this paper proposes diffusion latent beam search with estimatorwhich can select a better diffusion latent to maximize a given alignment reward at inference time.This paper point out that improving video quality with respect to alignment to prompts requires reward calibration by weighting existing metrics.1 IntroductionRecently, it has attracted a lot of attetion to 1) st..