When AI paints a new Rembrandt
- David Moreno
- 5 days ago
- 5 min read
Updated: 3 days ago
Can a painter who died more than 350 years ago return to us with a new work? That's what happened with The Next Rembrandt project, which analyzed the painter's works using AI to extract their "secrets of creation"...
In 2016, Delft University of Technology in the Netherlands, in partnership with Microsoft, unveiled a new Rembrandt painting produced using artificial intelligence. The software not only analyzed the painter's work to reveal his techniques, but also created an original 3D-printed portrait of a man in a black hat, depicting a stunning image.
Another example... The French collective Obvious created an original work using AI trained on a vast corpus of portraits produced since the Middle Ages: the Portrait of Edmond de Belamy . In October 2018, the piece was auctioned by Christie's in New York for the modest sum of $432,000.
The Next Rembrandt
This technological masterpiece is the result of a collaboration between art historians, developers, and data analysts that lasted over 18 months. While the choice to emulate the Dutch genius was no accident, as he is considered one of the greatest painters in art history, the challenge of reproducing the chiaroscuro that is his hallmark was significant. Introduced a century earlier in Italy by Caravaggio, the technique was perfected by Rembrandt in a unique way, through a skillful contrast between the darkness of the backgrounds and the light of the faces, which gives his paintings their distinctive depth.
The first step involved scanning over 160,000 fragments from 346 of the master's paintings in ultra-high definition. To capture the finest details, the analysis of Rembrandt's works was carried out using a deep learning algorithm to maximize the resolution and quality of the scans. The information from this invaluable collection was stored in a database, forming the digital foundation for creating the new Rembrandt.
Next, the theme of the painting had to be determined. Because the artist painted more portraits than any other subject, our team of engineers and academics focused on this theme during his most productive period, from 1632 to 1642. This study made it possible to outline the portrait of an imaginary character that Rembrandt could really have painted: " A Caucasian man, bearded, between 30 and 40 years old, wearing dark clothes, a ruff and a hat, looking to the right, " summarizes Emmanuel Flores, technical director of the project.
The third and most exciting step involved designing and developing an AI that learned the Flemish painter's style, particularly his mastery of chiaroscuro. Then, a facial recognition algorithm identified the geometric patterns Rembrandt used to paint the features of the human face. This software then applied the learned principles to create, "in its own way," eyes, a nose, a mouth, a beard, a forehead... Creating the man in the black hat required nearly 500 hours of computation!
Finally, to bring this 148-megapixel digital photograph to life, our team used 3D printing. An algorithm first identified the number of layers the Flemish master would have used had he painted The Next Rembrandt . Then, a printer deposited (layer by layer) onto the canvas the paint droplets necessary to control the colors. The result is astonishing, as the final rendering closely resembles an oil painting fresh from the 17th-century artist's studio.
Rest assured, though: while it's impossible for a layperson to distinguish this work from an original Rembrandt, an expert wouldn't be fooled... " Rembrandt would certainly have made variations on this. He would have used dots of paint, and in particular a distinct dot of paint on the nose to make it stand out. Not the smooth finish on the nose in this painting, which looks good, but doesn't really highlight it ," says the curator of the Rembrandt Museum in Amsterdam.
In short, this work is far "too perfect" to have been produced by an artist who often lived on the edge, and whose dynamic and tormented style has captivated minds for nearly four centuries. The national poet Joost van den Vondel said of Rembrandt that he was " the friend and son of the shadow, like the night owl ," a description that ill suits the affable and uncomplicated nature of this man in the black hat. Even if the master's touch is clearly present, the artist's soul is missing. The achievement is more technological than artistic.
The Portrait of Edmond de Belamy
The Portrait of Edmond de Belamy boasts the distinction of being the first original work produced by artificial intelligence and sold for several hundred thousand dollars. Let's explore this mysterious portrait, reminiscent of a notable figure straight out of a Maupassant novel, painted by a contemporary artist...
At the beginning of this adventure, three childhood friends, roommates in Paris, wondered how to democratize creation using artificial intelligence. They then created the Obvious collective and used software that had been placed in the public domain, which they fed with 15,000 portraits created between the 14th and 20th centuries.
The computer application is based on a recent artificial intelligence technique: Generative Adversarial Networks (GANs), developed by Canadian student Ian Goodfellow in 2014. The principle of a GAN is simple and ingenious: two neural networks are placed in competition. The first network is the Generator (or the forger, more appropriate here), while the second, its opponent, is the Discriminator (or the expert).
The expert analyzes each portrait produced by the forger and must determine whether it was created by a machine or a human. If the expert determines that the portrait was created by a machine, they disqualify it and inform the forger of how they detected that the work originated from a computer. This allows the forger to avoid repeating the mistake and thus improve their technique. This back-and-forth between the two opposing neural networks is repeated until the expert is no longer able to determine whether the painting was created by a computer or a human. This is how the Portrait of Edmond de Belamy came to be.
Needless to say, this sale sparked considerable debate. Nevertheless, these AI-based experiments may represent a turning point in the world of art. A new art form could emerge, much like photography or cinema, which developed thanks to technological inventions. According to Richard Lloyd, a Christie's executive, " Artists [...] will embrace artificial intelligence and make its effects felt in many ways, both strange and magnificent. "
However, artists will need to be more engaged, that is, more in touch with the very heart of the creative process of their works. For we can ask ourselves to what extent Goodfellow's algorithm and the data extracted from the 15,000 portraits are not ultimately the true authors of the Portrait of Edmond de Belamy ... Moreover, it is no coincidence that the painting is signed with the mathematical formula that gave rise to it and that "bel ami" is ultimately a good translation of "good fellow"!
This text is the result of an automated translation. The first version of this text (in French) is available at the following address:
Sources :
Site officiel du projet The Next Rembrandt
Un « nouveau » Rembrandt conçu par les algorithmes et imprimé en 3D
« The Next Rembrandt » : l'IA peut-elle créer des œuvres d'Art
The Next Rembrandt : quand un algorithme imite un grand maître
Rembrandt – Wikipedia
Demain, l'art sans artiste ?
Un tableau produit par intelligence artificielle vendu chez Christie's plus de 40 fois son estimation
Et l'intelligence s'invite dans l'art
Le Portrait d'Edmond de Belamy – Wikipedia
Le site officiel du collectif Obvious
