Style Decomposition and Content Preservation for Artistic Style Transfer

ICLR 2026 Conference Submission25038 Authors

Keywords: Style Transfer, Decompsing, Diffusion

Abstract:

Artistic style transfer is a crucial task that aims to transfer the artistic style of a style image to a content image, generating a new image with preserved content and a distinct style. With the advancement of image generation methods, significant progress has been made in artistic style transfer. However, the existing methods face two key challenges: i) style ambiguity, due to inadequate definition of style, making it difficult to transfer certain style attributes; ii) content nonrestraint, the lack of effective constraint information causes stylistic features of the content, such as color and texture, to seriously influence content preservation effectiveness.
To address this challenges, improving the quality of style transfer while ensuring effective content preservation, we propose SDCP, Style Decomposition and Content Preservation for Artistic Style Transfer, to achieve effective style transfer through style decomposition and content preservation. First, distinguishing from previous work, we propose a style decomposing module that effectively represents style based on three basic attributes (brushstrokes, color, and texture) enabling clear style definition. Second, we design a content preserving module that employs line drawings as constraints to discard style elements while preserving content, utilizing cross-modal alignment to preserving semantic. Finally, all representations are injected into the denoising U-Net through a conditional injection mechanism. Quantitative and qualitative experiments are conducted to demonstrate that SDCP outperforms the current state-of-the-art models.

Primary Area: applications to computer vision, audio, language, and other modalities

Submission Number: 25038