RMFlow: Refined Mean Flow by a Noise-Injection Step for Multimodal Generation

ICLR 2026 Conference Submission8367 Authors

Keywords: Mean Flow, Flow Matching, Noise-injection, Likelihood Maximization, Multimodal Generation

TL;DR: We RMFlow, an efficient multimodal generative model that integrates a coarse 1-NFE MeanFlow transport with a subsequent tailored noise-injection refinement step.

Abstract: Mean flow (MeanFlow) enables efficient, high-fidelity image generation, yet its single-function evaluation (1-NFE) generation often cannot yield compelling results. We address this issue by introducing RMFlow, an efficient multimodal generative model that integrates a coarse 1-NFE MeanFlow transport with a subsequent tailored noise-injection refinement step. RMFlow approximates the average velocity of the flow path using a neural network trained with a new loss function that balances minimizing the Wasserstein distance between probability paths and maximizing sample likelihood. RMFlow achieves competitive, often (near) state-of-the-art results on text-to-image, context-to-molecule, and time-series generation using 1-NFE, at a comparable computational cost to the baseline MeanFlows.

Primary Area: generative models

Submission Number: 8367