SuperFlow: Training Flow Matching Models with RL on the Fly

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Chen, Kaijie, Xu, Zhiyang, Shen, Ying, Lin, Zihao, Yao, Yuguang, Huang, Lifu
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866914247116587008
author Chen, Kaijie
Xu, Zhiyang
Shen, Ying
Lin, Zihao
Yao, Yuguang
Huang, Lifu
author_facet Chen, Kaijie
Xu, Zhiyang
Shen, Ying
Lin, Zihao
Yao, Yuguang
Huang, Lifu
contents Recent progress in flow-based generative models and reinforcement learning (RL) has improved text-image alignment and visual quality. However, current RL training for flow models still has two main problems: (i) GRPO-style fixed per-prompt group sizes ignore variation in sampling importance across prompts, which leads to inefficient sampling and slower training; and (ii) trajectory-level advantages are reused as per-step estimates, which biases credit assignment along the flow. We propose SuperFlow, an RL training framework for flow-based models that adjusts group sizes with variance-aware sampling and computes step-level advantages in a way that is consistent with continuous-time flow dynamics. Empirically, SuperFlow reaches promising performance while using only 5.4% to 56.3% of the original training steps and reduces training time by 5.2% to 16.7% without any architectural changes. On standard text-to-image (T2I) tasks, including text rendering, compositional image generation, and human preference alignment, SuperFlow improves over SD3.5-M by 4.6% to 47.2%, and over Flow-GRPO by 1.7% to 16.0%.
format Preprint
id arxiv_https___arxiv_org_abs_2512_17951
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle SuperFlow: Training Flow Matching Models with RL on the Fly
Chen, Kaijie
Xu, Zhiyang
Shen, Ying
Lin, Zihao
Yao, Yuguang
Huang, Lifu
Computer Vision and Pattern Recognition
Recent progress in flow-based generative models and reinforcement learning (RL) has improved text-image alignment and visual quality. However, current RL training for flow models still has two main problems: (i) GRPO-style fixed per-prompt group sizes ignore variation in sampling importance across prompts, which leads to inefficient sampling and slower training; and (ii) trajectory-level advantages are reused as per-step estimates, which biases credit assignment along the flow. We propose SuperFlow, an RL training framework for flow-based models that adjusts group sizes with variance-aware sampling and computes step-level advantages in a way that is consistent with continuous-time flow dynamics. Empirically, SuperFlow reaches promising performance while using only 5.4% to 56.3% of the original training steps and reduces training time by 5.2% to 16.7% without any architectural changes. On standard text-to-image (T2I) tasks, including text rendering, compositional image generation, and human preference alignment, SuperFlow improves over SD3.5-M by 4.6% to 47.2%, and over Flow-GRPO by 1.7% to 16.0%.
title SuperFlow: Training Flow Matching Models with RL on the Fly
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2512.17951