# Shortcutting Pre-trained Flow Matching Diffusion Models is Almost Free Lunch

Article · Xu Cai, Yang Wu, Qianli Chen, Haoran Wu, Lichuan Xiang, Hongkai Wen

- Archive id: arxiv:2510.17858
- Published: 2025-10-15
- Venue: neural information processing systems
- DOI: https://doi.org/10.52202/085713-5119
- arXiv: https://arxiv.org/abs/2510.17858
- Citations: 0
- Source URL: https://arxiv.org/abs/2510.17858
- PDF: https://arxiv.org/pdf/2510.17858
- Provenance: arxiv

## Abstract

We present an ultra-efficient post-training method for shortcutting large-scale pre-trained flow matching diffusion models into efficient few-step samplers, enabled by novel velocity field self-distillation. While shortcutting in flow matching, originally introduced by shortcut models, offers flexible trajectory-skipping capabilities, it requires a specialized step-size embedding incompatible with existing models unless retraining from scratch$\\unicode{x2013}$a process nearly as costly as pretraining itself. Our key contribution is thus imparting a more aggressive shortcut mechanism to standard flow matching models (e.g., Flux), leveraging a unique distillation principle that obviates the need for step-size embedding. Working on the velocity field rather than sample space and learning rapidly from self-guided distillation in an online manner, our approach trains efficiently, e.g., producing a 3-step Flux less than one A100 day. Beyond distillation, our method can be incorporated into the pretraining stage itself, yielding models that inherently learn efficient, few-step flows without compromising quality. This capability also enables, to our knowledge, the first few-shot distillation method (e.g., 10 text-image pairs) for dozen-billion-parameter diffusion models, delivering state-of-the-art performance at almost free cost.

_Abstract reproduced under CC0-1.0 (arXiv descriptive metadata)._

Topics: [Generative & Multimodal Models](https://space-kitty.com/topics/generative-multimodal)

---
Space Kitty Archive · https://space-kitty.com · API: https://space-kitty.com/api/v1 · MCP: https://space-kitty.com/mcp · Premium: $0.20/request via x402 (Base Sepolia (testnet)). Metadata: arXiv (CC0), OpenAlex (CC0), Europe PMC (CC BY/CC0 works), ClinicalTrials.gov.