From Reflection to Perfection: Scaling Inference-Time Optimization for Text-to-Image Diffusion Models via Reflection Tuning

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Zhuo, Le, Zhao, Liangbing, Paul, Sayak, Liao, Yue, Zhang, Renrui, Xin, Yi, Gao, Peng, Elhoseiny, Mohamed, Li, Hongsheng
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909588583874560
author Zhuo, Le
Zhao, Liangbing
Paul, Sayak
Liao, Yue
Zhang, Renrui
Xin, Yi
Gao, Peng
Elhoseiny, Mohamed
Li, Hongsheng
author_facet Zhuo, Le
Zhao, Liangbing
Paul, Sayak
Liao, Yue
Zhang, Renrui
Xin, Yi
Gao, Peng
Elhoseiny, Mohamed
Li, Hongsheng
contents Recent text-to-image diffusion models achieve impressive visual quality through extensive scaling of training data and model parameters, yet they often struggle with complex scenes and fine-grained details. Inspired by the self-reflection capabilities emergent in large language models, we propose ReflectionFlow, an inference-time framework enabling diffusion models to iteratively reflect upon and refine their outputs. ReflectionFlow introduces three complementary inference-time scaling axes: (1) noise-level scaling to optimize latent initialization; (2) prompt-level scaling for precise semantic guidance; and most notably, (3) reflection-level scaling, which explicitly provides actionable reflections to iteratively assess and correct previous generations. To facilitate reflection-level scaling, we construct GenRef, a large-scale dataset comprising 1 million triplets, each containing a reflection, a flawed image, and an enhanced image. Leveraging this dataset, we efficiently perform reflection tuning on state-of-the-art diffusion transformer, FLUX.1-dev, by jointly modeling multimodal inputs within a unified framework. Experimental results show that ReflectionFlow significantly outperforms naive noise-level scaling methods, offering a scalable and compute-efficient solution toward higher-quality image synthesis on challenging tasks.
format Preprint
id arxiv_https___arxiv_org_abs_2504_16080
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle From Reflection to Perfection: Scaling Inference-Time Optimization for Text-to-Image Diffusion Models via Reflection Tuning
Zhuo, Le
Zhao, Liangbing
Paul, Sayak
Liao, Yue
Zhang, Renrui
Xin, Yi
Gao, Peng
Elhoseiny, Mohamed
Li, Hongsheng
Computer Vision and Pattern Recognition
Recent text-to-image diffusion models achieve impressive visual quality through extensive scaling of training data and model parameters, yet they often struggle with complex scenes and fine-grained details. Inspired by the self-reflection capabilities emergent in large language models, we propose ReflectionFlow, an inference-time framework enabling diffusion models to iteratively reflect upon and refine their outputs. ReflectionFlow introduces three complementary inference-time scaling axes: (1) noise-level scaling to optimize latent initialization; (2) prompt-level scaling for precise semantic guidance; and most notably, (3) reflection-level scaling, which explicitly provides actionable reflections to iteratively assess and correct previous generations. To facilitate reflection-level scaling, we construct GenRef, a large-scale dataset comprising 1 million triplets, each containing a reflection, a flawed image, and an enhanced image. Leveraging this dataset, we efficiently perform reflection tuning on state-of-the-art diffusion transformer, FLUX.1-dev, by jointly modeling multimodal inputs within a unified framework. Experimental results show that ReflectionFlow significantly outperforms naive noise-level scaling methods, offering a scalable and compute-efficient solution toward higher-quality image synthesis on challenging tasks.
title From Reflection to Perfection: Scaling Inference-Time Optimization for Text-to-Image Diffusion Models via Reflection Tuning
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2504.16080