acceptodds
Desk rejected at ICLR 2027

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

Desk rejected at ICLR 2027, so left out of the paper lists and the map.

22 views

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, recent work, 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. We apply the DiffusionNFT objective to this predictor, making reward optimization well-defined for MeanFlow. Sampling remains based on the average velocity, preserving MeanFlow's fast few-step generation. We further prove that MeanFlowNFT inherits DiffusionNFT's strict policy-improvement guarantee. Experiments on image and video generation show that MeanFlowNFT consistently improves baselines. Moreover, it outperforms prior state-of-the-art RL-tuned few-step generators on most metrics ( of on SD3.5-M), and can even surpass multi-step RL-tuned diffusion while using only a few sampling steps. For instance, on Wan 2.1, -step MeanFlowNFT reaches a VBench score of , surpassing -step LongCat-Video RL ().

Primary area:
generative models
Keywords:
diffusion models, few-step generator, reinforcement learning, meanflow

Related papers

Discussion (0)

Sign in to comment.