🖼 AI Restores Paintings in Hours Instead of Months
A researcher from MIT has proposed a method for restoring paintings using AI. To demonstrate the approach, graduate student Alex Kachkine restored a painting by an unknown 15th-century Dutch artist. Over the centuries, the artwork has suffered significant damage: its surface was covered with cracks, and many fragments were lost entirely (see the video ⤴️).
Kachkine scanned the painting and restored it in several steps:
🔵 He created a restored "digital copy": algorithms filled in minor cracks based on nearby colors and reconstructed patterns.
🔵 Some elements were manually retouched by Kachkine in Photoshop. For example, deep cracks or missing faces were recreated using elements from other works by the artist. He avoided using generative AIs to preserve authenticity.
🔵 Finally, the algorithm transferred the restored elements onto a separate layer, which was printed on a thin transparent polymer and overlaid onto the original painting.
The film is invisible to the naked eye. Still, it can be easily removed without damaging the painting, allowing for future restorations. Applying the film took just 3.5 hours, whereas traditional restoration would have taken over 230 hours.
According to Kachkine, this method is suitable for relatively low-value paintings, which often lack resources for traditional restoration techniques.
🔐 Currently, about 70% of museum paintings are kept behind closed doors due to damage—thanks to digital restoration, these works could finally be accessible to the public.
