At the Massachusetts Institute of Technology, researcher Alex Kachkine introduced a new approach to restoring damaged oil paintings using generative AI. Instead of months of manual work, the model allows for digitally restoring lost fragments of the canvas in just a few hours, which can significantly speed up the restoration process.
During the demonstration of the method on the painting “Prado Adoration” by a master from the late fifteenth century, the researcher first scanned the image to identify over five thousand damaged areas. Using Adobe Photoshop, a digital mask was created to restore lost colors and patterns, while missing details — such as the face of the infant — were borrowed from other works by the same author.
After completing the digital reconstruction, the restored image was printed on a transparent polymer film, which was then carefully placed over the original canvas. In total, over fifty-seven thousand colors were used to fill the damaged areas. If necessary, the film can be easily removed without harming the original.
The method is suitable for paintings covered with varnish and smooth enough to ensure a tight fit of the polymer film. According to the researchers, this solution opens up new possibilities for museums and galleries, allowing for quick and affordable restoration of works that previously remained neglected due to the high cost of traditional restoration. This can help make more damaged paintings accessible to the public.
