Researchers at Chinese universities, backed by AI large model-powered restoration technology, are capable of reviving faded ancient Chinese paintings to somewhat bring back to life the original brilliance of precious cultural relics.
At the Laboratory of Art and Archaeology Image at the Zhejiang University in east China's Zhejiang Province, an artwork restoration team is employing advanced AI models to digitally restore the colors of the thousand-year-old traditional painting "Palace Lady Waving a Fan", which has significantly faded over time.
"By incorporating all our previous know-how, methods, paper outcomes about art restoration into a single AI model, we have obtained a very useful AI-powered restoration assistant," said Tang Tan, a research fellow with the School of Art and Archaeology of Zhejiang University.
In the past, restoring ancient paintings would mainly depend on the experience of restorers, who analyze the colors and conduct multiple digital simulations. Now, with the help of large AI models, the pigment composition of the painting can be analyzed, and AI can also factor in the painting's creation period and the artist's style to select the most accurate color scheme.
After the AI completes the visual simulation, restorers can manually verify the effects in the lab in the simulative color restoration.
"This is our materials lab. To restore the original colors of ancient paintings, we seek out materials to replicate the pigments historically used in the creation of artworks. For example, we apply these pigments onto reproductions of silk or paper, mimicking the original techniques and materials," Tang said.
The restoration team has digitally restored over 100 ancient paintings so far using the AI large model.
Moreover, they are collecting data on various colors, compositions and other aspects from past artworks, and using the vast information to retrain their AI model.
"We aim to integrate comprehensive knowledge of the full history of Chinese painting, including detailed depiction and color information of each artwork. This will enable us to build a database that will, in turn, enhance the expertise of our general AI model," Tang noted.
Faded ancient Chinese paintings revived with AI-powered restoration
