This startup uses AI to recover art buried under well-known paintings

When creating The Crouching Beggar back in 1902, Picasso painted over art by one of his earliest influences, Santiago Rusiñol of the Catalan modernisme movement. Now, British startup Oxia Palus has brought Rusiñol’s painting back to life by mapping it with machine learning and 3D printing it on canvas.

Founders George Cann and Anthony Bourached met as PhD students at University College London, focusing respectively on infrared spectroscopy applied to the Mars atmosphere and machine learning. Cann was already experimenting with Neural Style Transfer algorithms to make his own artwork when he met Bourached, and the two soon started thinking about how they could use AI to shape the future of art.

“There’s a very significant amount of art that gets lost,” Bourached says. “It was actually quite common with old masters to paint over work they did. For instance, it is estimated that around 30% of Van Gogh’s paintings have underpaintings by him, and being very poor during his Blue Period, Picasso used to paint over a lot of his work.”

Bourached and Cann first wrote a paper in 2019 with their findings. They titled it “Raiders of the Lost Art”, and despite it being just a proof of concept, it quickly went viral. MIT Technology Review covered it, mentioning that while the team’s outline-based images provided important insights into artists’ work and progression, “a way to reconstruct the lost painting more realistically,” including its colour and style, would have added unprecedented value.

Now, Oxia Palus has recreated Rusiñol’s style to generate plausible brushstrokes and colours when 3D printing his painting portraying Barcelona’s Parc del Laberint d’Horta.

“We realised this is actually an extremely hard problem, because you’re trying to layer something on canvas, which is a non-rigid layer, and there are only a few printers in the world that can do that,” Bourached says. “We’re one of the first to solve it.”

The team chose to recreate Rusiñol’s work because of the mystery surrounding it. “Rusiñol was about 20 years older than Picasso and a friend of his,” Bourached says. “It’s not clear why Picasso would have painted over his work.” Picasso’s The Crouching Beggar dates back to when the artist was living in poverty in Paris, and might represent a rejection of Rusiñol’s fascination with neoclassical landscapes.

The two co-founders 3D printed 100 canvases, available for purchase from Silicon Valley-based MORF Gallery for $11,111.11 each. In order to protect them from fraud or replication, the physical works are embedded with a code linked to a non-fungible token (NFT) unique to each piece.

Ultimately, Oxia Palus hopes to use AI to understand the past better and recover what’s been lost. The goal is to build relationships with museums — “there are hundreds to thousands of paintings with paintings underneath,” says Bourached — and improve the way hidden art is discovered and analysed, but also use their technology to enhance our understanding of dreams.

Follow their journey on Conception X cohort 4 @conceptionxtech and on