Flow Straighter and Faster: Efficient One-Step Generative Modeling via MeanFlow on Rectified Trajectories

2025-11-28

AI Summary

Existing one-step flow-based generative models face a fundamental trade-off: Rectified Flow requires multiple reflow iterations to approximate straight trajectories, incurring high computational cost; MeanFlow enables one-step generation but suffers from slow convergence and noise-induced supervision bias when trained on curved flows. This paper proposes Rectified MeanFlow, a novel framework that directly models the mean velocity field over rectified trajectories, integrating one-step rectification with truncated heuristic optimization. Our approach significantly suppresses trajectory curvature and mitigates training noise, enabling high-fidelity one-step generation after a single rectification—without iterative distillation or ODE solving. Extensive experiments on ImageNet at resolutions of 64×64, 256×256, and 512×512 demonstrate state-of-the-art performance among one-step flow methods, achieving superior sample quality while maintaining high training efficiency.

Technology Category

• Machine Learning: Deep Generative Models & Autoencoders
• Computer Vision: Diffusion Models for Vision
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

Flow-based generative models have recently demonstrated strong performance, yet sampling typically relies on expensive numerical integration of ordinary differential equations (ODEs). Rectified Flow enables one-step sampling by learning nearly straight probability paths, but achieving such straightness requires multiple computationally intensive reflow iterations. MeanFlow achieves one-step generation by directly modeling the average velocity over time; however, when trained on highly curved flows, it suffers from slow convergence and noisy supervision. To address these limitations, we propose Rectified MeanFlow, a framework that models the mean velocity field along the rectified trajectory using only a single reflow step. This eliminates the need for perfectly straightened trajectories while enabling efficient training. Furthermore, we introduce a simple yet effective truncation heuristic that aims to reduce residual curvature and further improve performance. Extensive experiments on ImageNet at 64, 256, and 512 resolutions show that Re-MeanFlow consistently outperforms prior one-step flow distillation and Rectified Flow methods in both sample quality and training efficiency. Code is available at https://github.com/Xinxi-Zhang/Re-MeanFlow.

Problem

• Reduces expensive numerical integration in flow-based generative models
• Eliminates need for multiple reflow iterations to straighten trajectories
• Improves training efficiency and sample quality for one-step generation

Innovation

• Rectified MeanFlow models mean velocity on rectified trajectories
• Uses single reflow step to avoid perfectly straightened paths
• Introduces truncation heuristic to reduce residual curvature