The Year in AI Papers
The year/Independent research

Paper 2604.08364

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

Junyao Gao, Sibo Liu, Jiaxing Li, Yanan Sun, Yuanpeng Tu, Fei Shen, Weidong Zhang, Cairong Zhao, Jun Zhang

Published
Apr 2026
Research lab
Independent
Citations
2
GitHub
131 stars

01 In brief

Summary

The paper introduces MegaStyle, a scalable data curation pipeline for constructing a large-scale, high-quality style dataset (MegaStyle-1.4M) with intra-style consistency and inter-style diversity.

It leverages the consistent text-to-image style mapping of large generative models like Qwen-Image to generate style pairs from curated prompts.

The pipeline includes image pool collection, prompt curation with balance sampling (yielding 170K style and 400K content prompts), and style image generation.

The authors propose style-supervised contrastive learning (SSCL) to fine-tune a style encoder (MegaStyle-Encoder) for reliable style similarity measurement, and train a FLUX-based style transfer model (MegaStyle-FLUX) using paired supervision.

Experiments show that MegaStyle-Encoder outperforms existing style encoders on retrieval benchmarks, and MegaStyle-FLUX achieves state-of-the-art style transfer performance with better text alignment and human preference scores compared to baselines.

Ablations confirm the importance of intra-style consistency and dataset quality.

The work contributes a scalable dataset, a style encoder, and a style transfer model, with future work aiming to scale the dataset to 10 million images.

02 From the paper

Abstract

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/.