Abstract
The goal of image style transfer is to render an image guided by a stylereference while maintaining the original content. Existing image-guided methodsrely on specific style reference images, restricting their wider applicationand potentially compromising result quality. As a flexible alternative,text-guided methods allow users to describe the desired style using textprompts. Despite their versatility, these methods often struggle withmaintaining style consistency, reflecting the described style accurately, andpreserving the content of the target image. To address these challenges, weintroduce FAGStyle, a zero-shot text-guided diffusion image style transfermethod. Our approach enhances inter-patch information interaction byincorporating the Sliding Window Crop technique and Feature Augmentation onGeodesic Surface into our style control loss. Furthermore, we integrate aPre-Shape self-correlation consistency loss to ensure content consistency.FAGStyle demonstrates superior performance over existing methods, consistentlyachieving stylization that retains the semantic content of the source image.Experimental results confirms the efficacy of FAGStyle across a diverse rangeof source contents and styles, both imagined and common.