Diffusion models have gained significant attention in low-level vision tasks due to their ability to generate high-quality and diverse samples. This survey provides a comprehensive overview of diffusion models applied in low-level vision, including theoretical foundations, practical applications, and future research directions. The paper introduces three diffusion modeling frameworks and explores their relationships with other deep generative models. It categorizes diffusion models based on underlying frameworks and target tasks, covering natural image processing, medical imaging, remote sensing, and video scenarios. The survey also summarizes commonly used benchmarks and evaluation metrics, and evaluates diffusion model-based techniques in three prominent tasks: image super-resolution, image deblurring, and low-light image enhancement. The paper identifies limitations of current diffusion models and proposes seven future research directions. The survey aims to provide a deep understanding of diffusion models in low-level vision tasks and offers a curated list of diffusion model-based techniques, datasets, and related information for over 20 low-level vision tasks. Diffusion models are characterized by a forward diffusion process that introduces noise and a reverse denoising process for image generation. They are typically classified into three subcategories: denoising diffusion probabilistic models (DDPMs), noise-conditional score networks (NCSNs), and stochastic differential equations (SDEs). DDPMs are known for their straightforward algorithmic flow and ease of integrating conditional controls, while NCSNs and SDEs are subject to detailed mathematical analysis. The paper discusses the theoretical foundations of diffusion models, their applications in low-level vision tasks, and their comparison with other deep generative models such as variational autoencoders (VAEs), normalizing flows (NFs), and generative adversarial networks (GANs). The survey highlights the strengths and limitations of diffusion models in low-level vision tasks and proposes future research directions. The paper concludes that diffusion models have shown significant success in low-level vision tasks and offer promising potential for future research.Diffusion models have gained significant attention in low-level vision tasks due to their ability to generate high-quality and diverse samples. This survey provides a comprehensive overview of diffusion models applied in low-level vision, including theoretical foundations, practical applications, and future research directions. The paper introduces three diffusion modeling frameworks and explores their relationships with other deep generative models. It categorizes diffusion models based on underlying frameworks and target tasks, covering natural image processing, medical imaging, remote sensing, and video scenarios. The survey also summarizes commonly used benchmarks and evaluation metrics, and evaluates diffusion model-based techniques in three prominent tasks: image super-resolution, image deblurring, and low-light image enhancement. The paper identifies limitations of current diffusion models and proposes seven future research directions. The survey aims to provide a deep understanding of diffusion models in low-level vision tasks and offers a curated list of diffusion model-based techniques, datasets, and related information for over 20 low-level vision tasks. Diffusion models are characterized by a forward diffusion process that introduces noise and a reverse denoising process for image generation. They are typically classified into three subcategories: denoising diffusion probabilistic models (DDPMs), noise-conditional score networks (NCSNs), and stochastic differential equations (SDEs). DDPMs are known for their straightforward algorithmic flow and ease of integrating conditional controls, while NCSNs and SDEs are subject to detailed mathematical analysis. The paper discusses the theoretical foundations of diffusion models, their applications in low-level vision tasks, and their comparison with other deep generative models such as variational autoencoders (VAEs), normalizing flows (NFs), and generative adversarial networks (GANs). The survey highlights the strengths and limitations of diffusion models in low-level vision tasks and proposes future research directions. The paper concludes that diffusion models have shown significant success in low-level vision tasks and offer promising potential for future research.