June 14, 2025

AI-Based Method Restores Priceless Renaissance Art in Under 4 Hours Rather Than Months

AI-Based Method Restores Priceless Renaissance Art In Under 4 Hours Rather Than Months
Before and after showing how the mask restored a damaged painting. Credit: Alex Kachkine

Somewhere in Cambridge at a MIT lab, a 15th-century painting depicting a Renaissance infant, once thought too damaged to display, now hangs fully restored. But this transformation didn’t come from a conservator’s brush — it came from a thin, transparent mask printed with more than 57,000 colors, carefully aligned and laid over the original work.

The restoration was made possible by a new method developed by MIT researcher Alex Kachkine. Instead of physically touching the painting, his technique digitally reconstructs the missing or damaged areas and prints that reconstruction onto a flexible film. The film is then applied directly to the painting’s surface, held in place with a removable conservation-grade varnish.

“It followed years of effort to try to get the method working,” Kachkine told The Guardian. “There was a fair bit of relief that finally this method was able to reconstruct and stitch together the surviving parts of the painting.”

This approach offers a radically faster and reversible alternative to traditional restoration — bringing new hope to the thousands of artworks locked away in museum storage, too damaged or too costly to restore by hand.

Pixel by Pixel Art Restoration

Image showing the layers required to recreate the artwork digitally without damaging the original - AI-Based Method Restores Priceless Renaissance Art In Under 4 Hours Rather Than Months
The original painting remains uncovered by the mask. Credit: Alex Kachkine

The painting in question is an oil-on-panel attributed to the Master of the Prado Adoration, a Netherlandish artist active in the late 1400s. It depicts the Adoration of the Magi and had been split into four panels, each worn and riddled with damage. Some 5,612 sections needed repair.

Ordinarily, restoring such a work would take months. Kachkine did it in under four hours.

He began by scanning the painting at high resolution. Then, he used digital inpainting tools, Photoshop edits, and machine-learning-assisted pattern transfers, he reconstructed what the painting might once have looked like. A baby’s face — missing entirely — was borrowed from a similar work by the same artist, The Presentation in the Temple, held at the National Gallery of Art in Washington.

The restored image served as a blueprint for a thin, transparent polymer “mask.” Using precision inkjet and laser printing, Kachkine laid down color and white pigment layers with microscopic accuracy — down to 42 microns per pixel — onto flexible sheets. These were then varnished and manually aligned onto the surface of the painting.

<!– Tag ID: zmescience_300x250_InContent_3

[jeg_zmescience_ad_auto size=”__300x250″ id=”zmescience_300x250_InContent_3″]

–>

The result is an overlay that brings the damaged sections visually back to life without touching the original paint. “It’s like applying a bandage that looks like the skin underneath,” Kachkine said.

Mapping the damaged areas of the painting to reconstruct with AI and painting tools - AI-Based Method Restores Priceless Renaissance Art In Under 4 Hours Rather Than Months
The first step in the process is mapping the damage of the painting. Credit: Alex Kachkine.

Why does this matter? Because more than 70% of paintings in major museum collections are in storage — many of them too damaged, or too obscure, to be prioritized for conservation.

“The method is likely to be most applicable to paintings of relatively low value that would otherwise be housed behind closed doors,” said Hartmut Kutzke, a conservation scientist at the University of Oslo. “It could widen public access to art, bringing damaged paintings out of storage and in front of a new audience.”

Kachkine’s method is not a replacement for traditional conservation. It doesn’t clean or stabilize the paint. But it drastically reduces the time and cost required to visually restore a damaged work.

In all, Kachkine’s mask used 57,314 distinct colors across an area of 66,205 square millimeters — about the size of a sheet of legal paper. Some colors were drawn from neighboring areas of the same painting. Others were interpolated based on texture or borrowed, like the infant’s face, from related works. In total, the process took just 3.5 hours to apply — compared to the estimated 200 hours it would have taken using brushes.

The mask is fully reversible. It adheres via a conservation-grade varnish and can be peeled off or removed with solvents without damaging the painting underneath.

Between Innovation and Integrity

AI-Based Method Restores Priceless Renaissance Art In Under 4 Hours Rather Than Months
Closeup of before and after masking. Credit:  Alex Kachkine.

Still, not everyone is convinced. “Up until now, conservation has been executed by people — humans who have decades of experience and who bring a human touch,” said Julian Baumgartner, a fine-art conservator in Chicago. “Take that away and it fundamentally changes our relationship with the art.”

Margaret Holben Ellis, a conservator at New York University, raised concerns about curatorial oversight: “It hasn’t been vetted by the curators and the art historians and the conservators,” she told IEEE Spectrum. “You can’t tell from a photograph whether this looks like a placemat or a well-restored painting.”

Others fear job loss and misuse. “I can see that this will probably change our industry and put a lot of people out of business,” said Peggy Van Witt, a fine-art conservator in Florida. There’s also the risk that digital overlays could be abused to pass off fake restorations — or even fakes — as originals.

Kachkine is sensitive to these concerns. “This technique does not replace most of what conservators do,” he said. “Hopefully, they agree.”

Honoring A Balance

Conservation has long walked a tightrope between staying true to the original art and what’s actually technically possible. Since the 18th century, experts have debated how much of a painting should be “corrected,” and whether those corrections should be detectable.

Kachkine tried to honor that balance. Up close, the printed mask is not perfect. Some pixels are visible. Minor misalignments exist. But that, he argues, is part of the debate. A restoration should never be indistinguishable from the original.

“I wanted this to be a tool,” he said — not a shortcut or a replacement for human expertise.

Kutzke, the Norwegian conservation scientist, agrees — cautiously. “As far as we can see now there are no serious concerns when it comes to the safety of the painting,” he wrote in a commentary in Nature. “But we need to gain more experience over a longer period with different types of paintings.”

The mask now hangs over the original painting in Kachkine’s apartment. The infant’s face looks serene, fully integrated into the scene.

But the implications stretch far beyond one baby.

Museums around the world are struggling with bloated collections and ballooning costs. Entire vaults sit idle, stacked with damaged paintings deemed too expensive to fix and too obscure to display.

Kachkine’s work offers a compelling alternative. If a printed, reversible mask can bring these forgotten works back to life — without harming them, without altering them — then maybe we’ve found a way to unearth a hidden trove of cultural heritage.

The method and findings were detailed in the journal Nature.