Overcoming the Curvature Bottleneck in MeanFlow

TL;DR

This paper introduces Rectified MeanFlow, a novel approach that straightens generative trajectories to improve convergence and quality in one-step flow models, significantly reducing FID scores and increasing speed.

Contribution

It proposes a self-distillation method with velocity rectification and a curvature pruning heuristic to enhance mean-velocity estimation along straight paths.

Findings

FID improved from 30.9 to 8.6 with same training budget

Outperforms recent 2-rectified flow++ by 33.4% in FID

Achieves 26x faster inference speed

Abstract

MeanFlow offers a promising framework for one-step generative modeling by directly learning a mean-velocity field, bypassing expensive numerical integration. However, we find that the highly curved generative trajectories of existing models induce a noisy loss landscape, severely bottlenecking convergence and model quality. We leverage a fundamental geometric principle to overcome this: mean-velocity estimation is drastically simpler along straight paths. Building on this insight, we propose Rectified MeanFlow, a self-distillation approach that learns the mean-velocity field over a straightened velocity field, induced by rectified couplings from a pretrained model. To further promote linearity, we introduce a distance-based truncation heuristic that prunes residual high-curvature pairs. By smoothing the optimization landscape, our method achieves strong one-step generation performance. We improve the FID of baseline MeanFlow models from 30.9 to 8.6 under same training budget, and outperform the recent 2-rectified flow++ by 33.4% in FID while running 26x faster. Our work suggests that the difficulty of one-step flow generation stems partially from the rugged optimization landscapes induced by curved trajectories. Code is available at https://github.com/Xinxi-Zhang/Re-MeanFlow.