One-Step Generation
Deep Learning
Generative Models
The frontier of single-step generative models: from Flow Matching to Drifting Models
Between 2025 and 2026, methods that overcome the multi-step inference of diffusion models and Flow Matching to generate high-quality images with a single network evaluation (1-NFE) have been rapidly advancing. This series curates four papers driving this field, tracing the technical evolution from extensions of Flow Matching to entirely new paradigms.
Contents
- Overview
- MeanFlow: One-Step Generation via Average Velocity
- Transition Matching: A Unified Framework via Discrete-Time Markov Transitions
- Terminal Velocity Matching: Distribution-Level Guarantees via Terminal-Time Regularization
- Drifting Models: A New Paradigm via Distribution Evolution During Training
Paper List
| # | Paper | Authors | Date | License |
|---|---|---|---|---|
| 1 | Mean Flows for One-step Generative Modeling | Geng, Deng, Bai, Kolter, He | 2025-05 | CC BY 4.0 |
| 2 | Transition Matching: Scalable and Flexible Generative Modeling | Shaul, Singer, Gat, Lipman | 2025-06 | CC BY 4.0 |
| 3 | Terminal Velocity Matching | Zhou, Parger, Haque, Song | 2025-11 | CC BY 4.0 |
| 4 | Generative Modeling via Drifting | Deng, Li, Li, Du, He | 2026-02 | CC BY 4.0 |