This paper presents high-quality image synthesis using diffusion probabilistic models, a class of latent variable models inspired by nonequilibrium thermodynamics. The authors achieve state-of-the-art results on the unconditional CIFAR10 dataset with an Inception score of 9.46 and an FID score of 3.17. They also obtain sample quality comparable to ProgressiveGAN on the 256x256 LSUN dataset. The key contribution is a novel connection between diffusion models and denoising score matching with Langevin dynamics, leading to a simplified training objective. The models naturally admit a progressive lossy decompression scheme, which can be interpreted as a generalization of autoregressive decoding. The paper discusses the background of diffusion models, their implementation details, and experimental results, including sample quality, rate-distortion analysis, and connections to other models. The authors conclude by highlighting the potential of diffusion models in various applications and the broader impact of generative models.This paper presents high-quality image synthesis using diffusion probabilistic models, a class of latent variable models inspired by nonequilibrium thermodynamics. The authors achieve state-of-the-art results on the unconditional CIFAR10 dataset with an Inception score of 9.46 and an FID score of 3.17. They also obtain sample quality comparable to ProgressiveGAN on the 256x256 LSUN dataset. The key contribution is a novel connection between diffusion models and denoising score matching with Langevin dynamics, leading to a simplified training objective. The models naturally admit a progressive lossy decompression scheme, which can be interpreted as a generalization of autoregressive decoding. The paper discusses the background of diffusion models, their implementation details, and experimental results, including sample quality, rate-distortion analysis, and connections to other models. The authors conclude by highlighting the potential of diffusion models in various applications and the broader impact of generative models.