Deep generative models have made significant advancements, particularly with diffusion models addressing the limitations of generative models. A comprehensive review paper, co-authored by The Chinese University of Hong Kong, Westlake University, MIT, and others, has been published in IEEE TKDE, delving into the progress and applications of diffusion models. Techniques such as knowledge distillation, improved training methods, and accelerated pre-training models have enhanced the efficiency of diffusion models. These models are not only successfully applied in image generation but also capable of converting text into images and enabling editing functions, showcasing a robust technological application future.