Can A.I. Resurrect a Delacroix Mural That Was Destroyed in a Fire More Than 150 Years Ago?
A new project called Digital Delacroix is training cutting-edge technology on the French painter’s style to unravel the lost artwork’s secrets

In the final days of the Paris Commune of 1871, a radical experiment in self-governance and insurrection that started after the nation’s humiliating military loss to Prussia, the Communards realized that their cause had gone up in flames.
The French Army, initially pushed out of the city by the Communards, returned to Paris with a vengeance. In the mayhem, the Communards set fire to the city, including the Tuileries Palace, the Palais de Justice and the Hôtel de Ville, which served as the city hall of Paris and the seat of the Communard government.
Eventually, the army retook Paris, and the stately buildings were reconstructed. But the archives and art within the Hôtel de Ville were forever lost. Among the most devastating losses were a series of allegorical murals by the Romantic painter Eugène Delacroix.
At their center was Peace Descends to Earth, a vivid panel that 19th-century French critic Théophile Gautier described as “the earth weeping, raising her eyes to heaven to plead for an end to her sorrows,” per the New York Times’ Frank Rose. The only records of the mural’s past glory are a photograph, several sketches, a few etchings and a pair of watercolors that the French emperor Napoleon III gave to Queen Victoria in 1855.
Now, researchers are hoping to unravel the mural’s secrets. Virtually recreating Peace Descends to Earth with the help of artificial intelligence is among the goals of Digital Delacroix, an ambitious project led by Barthélémy Jobert, an art historian at Sorbonne University in Paris.
/https://tf-cmsv2-smithsonianmag-media.s3.amazonaws.com/filer_public/b9/48/b948e6d3-3e1a-4cf9-82cd-b9976b86da15/hotel_ville_paris_burnt_commune_1871.jpg)
As Jobert sees it, well-trained A.I. can help infer links between Delacroix’s existing studies for the mural and his distinctive style.
“We won’t give you an exact reproduction of the room as it was. That’s impossible,” Jobert tells the Times. “But we will give you what it could have been.”
Schmidt Sciences, a philanthropic venture by former Google CEO Eric Schmidt and his wife Wendy, announced this week that Digital Delacroix was the first recipient of its Humanities and A.I. Virtual Institute (HAVI) grants. The funding is in the high six figures, according to the Times.
/https://tf-cmsv2-smithsonianmag-media.s3.amazonaws.com/filer_public/03/3e/033e9c98-e94b-4e25-8856-e5d4c630425a/eugene_delacroix_-_sketch_for_peace_descends_to_earth_-_wga06217.jpg)
“We study humanities for much the same reason we study science: to gain knowledge of our world and our role within it,” Wendy Schmidt says in a statement. She explains that the initiative seeks out projects that marry “emerging technologies” with “deeply intertwined human histories and cultures.”
Jobert’s project fits the bill. In addition to the mural reconstruction, Digital Delacroix also promises to digitize some 2,500 of Delacroix’s letters and make murals painted at the Palais Bourbon and the Palais du Luxembourg—home to the French National Assembly and Senate, respectively—more accessible to the public.
Because letter digitization is a relatively straightforward process, the first order of business is to take high-resolution photos of the murals and, using photogrammetry, build virtual rooms that visitors can tour.
/https://tf-cmsv2-smithsonianmag-media.s3.amazonaws.com/filer_public/df/11/df1128b6-256e-44cd-b08c-9604722069c7/reconstitution-delacroix-2.jpg)
With these high-quality photos, Jobert and his team can then begin their secondary goal: determining which sections of the murals were painted by Delacroix himself and which were made by his assistants.
“This question of multiple authorship is a really tricky one,” Brent Seales, a computer scientist at the University of Kentucky and the director of HAVI, tells the Times.
Seales recently led the A.I.-assisted effort to decode 2,000-year-old scrolls charred by Mount Vesuvius’ eruption. Now, he hopes that A.I. will be able to discern faint stylistic differences between Delacroix and other artists.
“Evidence of authorship is subtle and individualized—a constellation of small things tied to the specific behavior and training of a unique person,” Seales tells Artnet’s Jo Lawson-Tancred. “While individual things might be a weak indicator, like a stroke style or a color selection, the aggregation of a number of these things makes for a unique signature.”
Technology that can identify different brushstrokes is experimental. Instead of generative A.I. like ChatGPT, which has advanced by leaps and bounds since its launch in 2022, this phase of the Digital Delacroix project relies on analytical A.I.
Progress in that field is slower. Still, “We think there’s a high possibility it will work,” Xavier Fresquet, deputy director of the Sorbonne Center for Artificial Intelligence, tells the Times.
Eventually, the researchers hope that the A.I. will become a Delacroix expert, able to instantly pick out his brushstrokes from existing works and subsequently determine how the painter might have crafted the murals in the Hôtel de Ville.
“A.I. is a new lens through which we can view master works,” Seales tells Artnet. “Anything that resurrects our appreciation of such a significant artist is a great reminder of the genius of the past and the promise of the future.”