Leap method speeds up diffusion model sampling with accuracy
LEAP: Accelerating ODE-based Diffusion Sampling with Optimal Linear Leaps
Computer Vision and Pattern Recognition
Summary
Diffusion models help computers create images and 3D content by gradually improving guesses. Usually, speeding this process up makes the final output worse when using rough time steps. The authors designed LEAP, a new technique that carefully predicts future steps without extra work, based on understanding the path these guesses take. This method works better and faster on many tasks like turning text into images or creating 3D models from images, and it can be easily added to existing tools.
What this means in practice
- •For ai application developers: Generate images from text faster without losing quality by integrating LEAP into diffusion model samplers.
- •For 3d content creators: Improve speed and accuracy when converting 2D images into 3D models by applying LEAP-enhanced diffusion sampling.
Authors
Haowei Zheng, Zhenyu Zhou, Jiawei Chen, Can Wang, Defang Chen
Abstract
Diffusion models primarily rely on general-purpose ODE-based numerical methods for training-free accelerated sampling, but performance degrades with coarse time discretization. In this paper, we propose optimal linear leaps (LEAP), a specialized method grounded in theoretical and empirical investigations of sampling trajectory geometry. Built upon low-dimensionality and cross-trajectory consistency, LEAP formulates a supervised least-squares optimization problem within the local trajectory subspace, achieving a precise approximation of future states without model evaluations. Extensive experiments demonstrate that LEAP outperforms previous methods across diverse modalities, including text-to-image (Stable Diffusion 3.5, FLUX.1-dev) and image-to-3D (TRELLIS.2) generation. Furthermore, LEAP can be seamlessly integrated into existing solver-based methods, offering a plug-and-play enhancement for modern diffusion models. Code is available at https://github.com/zju-pi/LEAP.