Given an arbitrary content and style image, arbitrary style transfer aims to render a new stylizedimage which preserves the content image's structure and possesses the style image's style. Existingarbitrary style transfer methods are based on either small models or pre-trained large-scale models.The small model-based methods fail to generate high-quality stylized images, bringing artifacts anddisharmonious patterns. The pre-trained large-scale model-based methods can generate high-qualitystylized images but struggle to preserve the content structure and cost long inference time. To thisend, we propose a new framework, called SPAST, to generate high-quality stylized images withless inference time. Specifically, we design a novel Local-global Window Size Stylization Module(LGWSSM)tofuse style features into content features. Besides, we introduce a novel style prior loss,which can dig out the style priors from a pre-trained large-scale model into the SPAST and motivatethe SPAST to generate high-quality stylized images with short inferencethis http URLconduct abundantexperiments to verify that our proposed method can generate high-quality stylized images and lessinference time compared with the SOTA arbitrary style transfer methods.
View on arXiv@article{zhang2025_2505.08695,
title={ SPAST: Arbitrary Style Transfer with Style Priors via Pre-trained Large-scale Model },
author={ Zhanjie Zhang and Quanwei Zhang and Junsheng Luan and Mengyuan Yang and Yun Wang and Lei Zhao },
journal={arXiv preprint arXiv:2505.08695},
year={ 2025 }
}