We present high quality image synthesis results using diffusion probabilistic models,a class of latent variable models inspired by considerations from nonequilibriumthermodynamics. Our best results are obtained by training on a weighted variationalbound designed according to a novel connection between diffusion probabilisticmodels and denoising score matching with Langevin dynamics, and our models nat-urally admit a progressive lossy decompression scheme that can be interpreted as ageneralization of autoregressive decoding. On the unconditional CIFAR10 dataset,we obtain an Inception score of 9.46 and a state-of-the-art FID score of 3.17. On256x256 LSUN, we obtain sample quality similar to ProgressiveGAN. Our imple-mentation is available at https://github.com/hojonathanho/diffusion.