Mural Restoration with Integration of Prompt and Edge Constraints: A Study on Generation Accuracy Optimization Based on Stable Diffusion and ControlNet

This study presents a mural restoration method leveraging Generative Adversarial Network (GAN) technology, specifically combining Stable Diffusion and ControlNet models. The method addresses key challenges in traditional mural restoration, such as loss of detail and color deviations, by enhancing the accuracy and quality of the restoration process. Initially, the Stable Diffusion model generates the mural images, followed by the use of ControlNet to control the image’s structure and edge information. This ensures that the restored murals stay true to the original style and details. By incorporating edge constraints and carefully designed textual Prompts, the method prevents distortions and style drifts, leading to more precise restoration results. Experimental results show that this approach preserves the original appearance of the murals more accurately than traditional techniques, offering restored images that align better with human visual perception in terms of naturalness and artistic quality. The study introduces a novel technical solution for mural restoration, enhancing the scope of digital restoration methods and providing valuable support for the protection and preservation of cultural heritage. It holds both academic and practical significance.