MeanFlowNFT: Bringing Forward-Process RL to Average-Velocity Generators

Forum topic · 小凯 · 2026-07-18

Summary

MeanFlowNFT is a new reinforcement learning framework that aligns MeanFlow generators with human preferences while preserving their fast few-step sampling capability. MeanFlow generators predict average velocities over time intervals, whereas efficient RL methods like DiffusionNFT optimize instantaneous velocities via a forward-process objective that requires no reverse trajectories or likelihood estimation. MeanFlowNFT bridges this gap by constructing an induced instantaneous-velocity predictor based on the MeanFlow identity, applying the DiffusionNFT objective to it, while sampling remains based on average velocity. The method inherits strict policy improvement guarantees. Experiments on image and video generation show consistent gains over baselines, beating prior state-of-the-art few-step RL-tuned generators on 6 of 8 metrics for SD3.5-M. On Wan 2.1, 4-step MeanFlowNFT achieves a VBench score of 84.33, surpassing the 50-step LongCat-Video RL (82.57), showing few-step models can exceed multi-step RL-tuned diffusion models.

Overview

Field: Computer Vision (CV)
Authors: Yushi Huang, Xiangxin Zhou, Jun Zhang, Liefeng Bo, Tianyu Pang
Published: 2026-07-16
arXiv: 2607.15273

Abstract (translated)

MeanFlow generators achieve fast few-step sampling by predicting average velocities over time intervals, making them attractive for efficient generation. Reinforcement learning (RL) has become a powerful way to align diffusion and flow models with human preferences and task-specific objectives. In particular, DiffusionNFT offers an efficient forward-process RL framework that does not require reverse-process trajectories or likelihood estimation. However, applying such RL methods to MeanFlow remains underexplored: DiffusionNFT optimizes instantaneous velocities, whereas MeanFlow samples with average velocities.

To bridge this gap, the authors introduce MeanFlowNFT. Inspired by the MeanFlow identity, which bridges average and instantaneous velocities, they construct an induced instantaneous-velocity predictor and apply the DiffusionNFT objective to it, making reward optimization for MeanFlow well-defined. Sampling still relies on average velocity, preserving MeanFlow's fast few-step generation advantage. They further prove that MeanFlowNFT inherits the strict policy improvement guarantees of DiffusionNFT.

Results

Experiments on image and video generation show that MeanFlowNFT consistently improves over baselines:

Outperforms prior state-of-the-art few-step RL-tuned generators on most metrics (6 of 8 metrics on SD3.5-M).
Surpasses even multi-step RL-tuned diffusion models using only a few sampling steps.
On Wan 2.1, 4-step MeanFlowNFT achieves a VBench score of 84.33, exceeding the 50-step LongCat-Video RL (82.57).

Original abstract

MeanFlow generators achieve fast few-step sampling by predicting average velocities over time intervals, making them attractive for efficient generation. Reinforcement learning (RL) has become a powerful way to align diffusion and flow models with human preferences and task-specific objectives. In particular, DiffusionNFT offers an efficient forward-process RL framework that does not require reverse-process trajectories or likelihood estimation. However, applying such RL methods to MeanFlow remains underexplored. DiffusionNFT optimizes instantaneous velocities, whereas MeanFlow samples with average velocities. To bridge this gap, we introduce MeanFlowNFT. Inspired by the MeanFlow identity, which bridges average and instantaneous velocities, we construct an induced instantaneous-velocity predictor...

*Auto-collected on 2026-07-18.*