From inpainting to painting: exploring conservation of Chinese paintings with generative artificial intelligence

Chinese painting conservation faces several challenges, such as the inherent conflict between the conservation principles of minimal intervention, recognizability, and reversibility (Muñoz-Viñas, 2012), and the traditional pursuit of completeness in restoration (Gao & Jones, 2021; Liszewska, 2015). Additionally, a letter survives between two influential Qing Dynasty painters, Shitao and Bada Shanren, requesting a painting, though the work is lost. Recently, Generative Artificial Intelligence (GenAI) has seen increasing applications in art-related fields. This study explores the potential of using GenAI for virtual restoration and reconstruction in the context of Chinese painting. The research employs a research-creation methodology, primarily using Stable Diffusion to fill in missing portions of Chinese paintings and to reconstruct a historically lost artwork. The results indicate that GenAI can assist in virtual restoration and to address some challenges in Chinese painting conservation. Furthermore, it supports to visualize lost paintings in the original artists' styles. The study offers insights into possibilities of GenAI for art conservation, exhibition and research.