Exploring and Exploiting Stability in Latent Flow Matching

2026-05-08

AI Summary

This work investigates how to maintain the generation quality and efficiency of Latent Flow Matching (LFM) while substantially reducing both training data requirements and model capacity. Through theoretical analysis, the authors demonstrate that the stability of LFM stems inherently from its flow-matching objective, enabling the development of a data-efficient training strategy and a coarse-to-fine inference framework that synergistically combines a lightweight and a high-capacity model. By integrating a sample scoring criterion with architectural compression, the proposed approach achieves significant reductions in training data volume without compromising generation fidelity across multiple datasets. Furthermore, it accelerates inference by more than twofold, markedly lowering annotation costs and computational overhead.

Abstract

In this work, we show that Latent Flow-Matching (LFM) models are robust to different types of perturbations, including data reduction and model capacity shrinkage. We characterize this stability by their tendency to generate similar outputs under identical noise seeds. We provide a perspective relating this phenomenon to flow matching theory, which indicates that this stability is inherent to the FM objective. We further exploit this stability to derive practical algorithms for more efficient training and inference. Concretely, first, we show that by training LFM models on significantly reduced datasets, the performance does not degrade perceptually or quantitatively. This yields multiple advantages, such as reducing training time by converging faster under limited compute budget, and alleviating annotation effort when training conditional models. Second, LFM stability under architectural shrinkage gives rise to a two-model coarse-to-fine approach, one using a light-weight architecture for the first phase of the FM trajectory, and one with higher capacity for the second, thereby reducing the inference cost substantially. To determine which samples are informative, we introduce three sample-scoring criteria and evaluate them under standard metrics for generative models. Our results are thoroughly evaluated on multiple datasets, demonstrating the practical advantage of this stability, including data saving and a more than two-fold inference speedup while generating comparable outputs.

Problem

• Latent Flow Matching
• training efficiency
• inference cost
• data reduction
• model compression

Innovation

• Latent Flow Matching
• stability
• data efficiency
• coarse-to-fine inference
• sample scoring