**Align Your Steps (AYS)** is a novel framework for optimizing sampling schedules in diffusion models, significantly enhancing the quality of outputs, especially in few-step synthesis. The framework leverages methods from stochastic calculus to find optimal schedules tailored to different solvers, trained DMs, and datasets. By minimizing an upper bound on the Kullback-Leibler divergence (KLUB) between the true and linearized generative SDEs, AYS optimizes the sampling schedule to improve output quality. The method is evaluated on various image, video, and 2D toy data synthesis benchmarks, showing superior performance compared to previous hand-crafted schedules. The optimized schedules are found to generalize well to popular ODE solvers and are applicable to all DMs regardless of data modality. The contributions of AYS include establishing the dependency of optimal sampling schedules on dataset characteristics, introducing a principled and general framework for optimization, improving upon existing heuristic schedules, and providing optimized schedules for commonly used models.**Align Your Steps (AYS)** is a novel framework for optimizing sampling schedules in diffusion models, significantly enhancing the quality of outputs, especially in few-step synthesis. The framework leverages methods from stochastic calculus to find optimal schedules tailored to different solvers, trained DMs, and datasets. By minimizing an upper bound on the Kullback-Leibler divergence (KLUB) between the true and linearized generative SDEs, AYS optimizes the sampling schedule to improve output quality. The method is evaluated on various image, video, and 2D toy data synthesis benchmarks, showing superior performance compared to previous hand-crafted schedules. The optimized schedules are found to generalize well to popular ODE solvers and are applicable to all DMs regardless of data modality. The contributions of AYS include establishing the dependency of optimal sampling schedules on dataset characteristics, introducing a principled and general framework for optimization, improving upon existing heuristic schedules, and providing optimized schedules for commonly used models.