While debates raged over whether artificial intelligence could replace connoisseurs and diagnosticians in the authentication of artworks, we didn’t see this coming. But AI has just entered the field of art restoration.
Just a few days ago, Nature published an article illustrating the first application of artificial intelligence in the physical restoration of a work of art: Physical restoration of a painting with a digitally constructed mask.
This isn’t about a robotic restorer carefully painting touch-ups, but rather a technique that, as a first step, uses already well-developed algorithms to fill in gaps in the digital image of the painted surface. AI models based on the latest transformers are then used to virtually reconstruct the larger missing details, like a portion of Baby Jesus’s head.
After all, the ability to predict missing parts—just like predicting the next word in Large Language Models—has recently advanced tremendously, achieving results we could hardly have imagined just a short time ago.
What’s new is that these digital images are now transformed into a physical film, where the colored areas are reconstructed and then affixed to the surface of the painting using a transparent varnish. Et voilà: the restoration is done —or at least the pictorial integration part—but achieved in a fraction of the time required by traditional “human” methods.
The first systems for virtual painting restoration—techniques aimed at providing a digitally “restored” image—were developed in the early 1990s at the Communications and Image Lab of the University of Florence, where I graduated. There, techniques were studied to simulate both cleaning (i.e., removal of incoherent material from the surface) and the filling of micro-cracks and losses.
Much has changed since then.
The results presented are impressive, and the author of the article suggests that this low-cost technique he developed could replace human restorers for many works that currently remain hidden from the public due to a lack of funding for restoration, and are therefore confined to storage.
But how does all this fit into the most recent theories of restoration?
If it’s true that, from the Athens Charter of 1931 to today, the approaches to restoration and its objectives have evolved—alongside the concept of authenticity that underpinned Cesare Brandi’s theory—this new automatic, predictive restoration introduces yet another layer to the discussion.
How do material alteration and the image changes that result from this new technique relate to the recurring principles in traditional restoration theory—such as authenticity, reversibility, and recognizability of interventions; the relationship between restorers and scientists; and the importance of studying the context and materials of the work?
Unless, we conclude—as Salvator Muñoz Viñas does in his recent Theory of Contemporary Restoration—that all these are outdated concepts.
