<scp>MGIRN</scp> : A Mask‐Guided Network for Layout‐Preserving and Creative Image Regeneration

ABSTRACT Preserving an image's layout while enabling creative and semantically consistent modifications to specific regions is a significant challenge in image regeneration. Existing methods often compromise layout fidelity to achieve content diversity. This paper introduces the mask‐guided image regeneration network (MGIRN), a framework leveraging a region localization layer and an information extraction module to regenerate the masked image while maintaining global layout integrity. Built on diffusion‐based generative models, MGIRN ensures high‐quality outputs by balancing structural preservation and semantic richness. We evaluate MGIRN using a newly curated interior design dataset with semantic masks. Experimental results demonstrate that MGIRN outperforms other methods in layout preservation, semantic coherence, and content diversity, offering a robust solution for interior design, advanced image editing, and other layout‐constrained applications.