MegaStyle: Constructing Diverse and Scalable Style Dataset via Consistent Text-to-Image Style Mapping

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Gao, Junyao, Liu, Sibo, Li, Jiaxing, Sun, Yanan, Tu, Yuanpeng, Shen, Fei, Zhang, Weidong, Zhao, Cairong, Zhang, Jun
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915945412296704
author Gao, Junyao
Liu, Sibo
Li, Jiaxing
Sun, Yanan
Tu, Yuanpeng
Shen, Fei
Zhang, Weidong
Zhao, Cairong
Zhang, Jun
author_facet Gao, Junyao
Liu, Sibo
Li, Jiaxing
Sun, Yanan
Tu, Yuanpeng
Shen, Fei
Zhang, Weidong
Zhao, Cairong
Zhang, Jun
contents In this paper, we introduce MegaStyle, a novel and scalable data curation pipeline that constructs an intra-style consistent, inter-style diverse and high-quality style dataset. We achieve this by leveraging the consistent text-to-image style mapping capability of current large generative models, which can generate images in the same style from a given style description. Building on this foundation, we curate a diverse and balanced prompt gallery with 170K style prompts and 400K content prompts, and generate a large-scale style dataset MegaStyle-1.4M via content-style prompt combinations. With MegaStyle-1.4M, we propose style-supervised contrastive learning to fine-tune a style encoder MegaStyle-Encoder for extracting expressive, style-specific representations, and we also train a FLUX-based style transfer model MegaStyle-FLUX. Extensive experiments demonstrate the importance of maintaining intra-style consistency, inter-style diversity and high-quality for style dataset, as well as the effectiveness of the proposed MegaStyle-1.4M. Moreover, when trained on MegaStyle-1.4M, MegaStyle-Encoder and MegaStyle-FLUX provide reliable style similarity measurement and generalizable style transfer, making a significant contribution to the style transfer community. More results are available at our project website https://jeoyal.github.io/MegaStyle/.
format Preprint
id arxiv_https___arxiv_org_abs_2604_08364
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle MegaStyle: Constructing Diverse and Scalable Style Dataset via Consistent Text-to-Image Style Mapping
Gao, Junyao
Liu, Sibo
Li, Jiaxing
Sun, Yanan
Tu, Yuanpeng
Shen, Fei
Zhang, Weidong
Zhao, Cairong
Zhang, Jun
Computer Vision and Pattern Recognition
In this paper, we introduce MegaStyle, a novel and scalable data curation pipeline that constructs an intra-style consistent, inter-style diverse and high-quality style dataset. We achieve this by leveraging the consistent text-to-image style mapping capability of current large generative models, which can generate images in the same style from a given style description. Building on this foundation, we curate a diverse and balanced prompt gallery with 170K style prompts and 400K content prompts, and generate a large-scale style dataset MegaStyle-1.4M via content-style prompt combinations. With MegaStyle-1.4M, we propose style-supervised contrastive learning to fine-tune a style encoder MegaStyle-Encoder for extracting expressive, style-specific representations, and we also train a FLUX-based style transfer model MegaStyle-FLUX. Extensive experiments demonstrate the importance of maintaining intra-style consistency, inter-style diversity and high-quality for style dataset, as well as the effectiveness of the proposed MegaStyle-1.4M. Moreover, when trained on MegaStyle-1.4M, MegaStyle-Encoder and MegaStyle-FLUX provide reliable style similarity measurement and generalizable style transfer, making a significant contribution to the style transfer community. More results are available at our project website https://jeoyal.github.io/MegaStyle/.
title MegaStyle: Constructing Diverse and Scalable Style Dataset via Consistent Text-to-Image Style Mapping
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2604.08364