Few-Step Generation via Data-Space Iteration

2026-10-08

AI Summary

This study addresses the limitation of existing few-step flow matching generation methods, which rely on fixed timestep discretization and consequently incur high tuning costs with poor adaptability. To overcome this, we propose a data-space iterative framework that abandons conventional flow-space iteration by directly refining predictions from noise within the data space. This approach eliminates dependence on probability flow discretization and removes the need for manual design or search of timestep schedules. Combined with Distribution Matching Distillation (DMD), our method employs a shared generator to conduct multi-round refinement training. On ImageNet 256×256, the proposed framework surpasses standard baselines and achieves performance comparable to search-optimized variants, demonstrating efficient schedule-free few-step generation.

Abstract

Flow matching has emerged as a scalable paradigm for training high-quality generative models, but sampling from the learned probability flow requires many network evaluations. Distillation can reduce this cost to one or a few evaluations; however, one-step generation often sacrifices quality, making few-step generation the practical operating regime. Existing few-step methods perform their iterative computation along the probability flow and therefore require a fixed, manually chosen timestep discretization. This discretization is often chosen heuristically and is expensive to tune; it may also be restrictive when refinement difficulty differs across samples or spatial locations. We introduce data-space iteration, a few-step generation framework that removes flow discretization altogether. Starting from noise, a shared generator directly refines its prediction in data space, with every iteration trained to produce the best sample permitted by its capacity. Our formulation integrates with distribution matching distillation (DMD) with minimal changes, enabling a controlled comparison between iteration methods under matched training settings. On class-conditional ImageNet 256x256, data-space iteration outperforms standard discretization baselines and matches or improves upon variants selected through schedule search, without requiring schedule-specific training. These results show that data-space iteration provides a simple and effective alternative to discretized flow-space iteration for fast generation.

Problem

• Few-step generation
• Flow matching
• Time discretization
• Generative models
• Distillation

Innovation

• Data-Space Iteration
• Few-Step Generation
• Flow Matching
• Distribution Matching Distillation
• Schedule-Free