Fast-DDPM is a fast denoising diffusion probabilistic model designed to improve the training and sampling speed of diffusion models while maintaining high-quality image generation. Unlike traditional DDPMs that use 1,000 time steps, Fast-DDPM uses only 10 time steps, significantly reducing training and sampling times. The model aligns training and sampling procedures to optimize time-step utilization, introducing two efficient noise schedulers with 10 time steps: one with uniform sampling and another with non-uniform sampling. Fast-DDPM was evaluated on three medical image-to-image generation tasks: multi-image super-resolution, image denoising, and image-to-image translation. It outperformed DDPM and other state-of-the-art methods in all tasks, reducing training time to 0.2× and sampling time to 0.01× compared to DDPM. Fast-DDPM is approximately 100 times faster during sampling and about 5 times faster during training. The model's efficiency allows for faster and more practical applications in medical imaging, improving diagnosis and treatment planning. The study highlights the importance of optimizing time-step utilization and selecting appropriate noise schedulers for different medical imaging tasks. Future work could explore adaptive mechanisms for dynamically adjusting time steps based on input complexity and task requirements.Fast-DDPM is a fast denoising diffusion probabilistic model designed to improve the training and sampling speed of diffusion models while maintaining high-quality image generation. Unlike traditional DDPMs that use 1,000 time steps, Fast-DDPM uses only 10 time steps, significantly reducing training and sampling times. The model aligns training and sampling procedures to optimize time-step utilization, introducing two efficient noise schedulers with 10 time steps: one with uniform sampling and another with non-uniform sampling. Fast-DDPM was evaluated on three medical image-to-image generation tasks: multi-image super-resolution, image denoising, and image-to-image translation. It outperformed DDPM and other state-of-the-art methods in all tasks, reducing training time to 0.2× and sampling time to 0.01× compared to DDPM. Fast-DDPM is approximately 100 times faster during sampling and about 5 times faster during training. The model's efficiency allows for faster and more practical applications in medical imaging, improving diagnosis and treatment planning. The study highlights the importance of optimizing time-step utilization and selecting appropriate noise schedulers for different medical imaging tasks. Future work could explore adaptive mechanisms for dynamically adjusting time steps based on input complexity and task requirements.