khoji parkAI · Paris · 2018
An AI-Generated Portrait Sold at Christie’s
Paris-based arts collective Obvious created a portrait using a generative adversarial network algorithm.
Read in slides1. Creation by Collective
The Paris-based arts collective Obvious, consisting of Hugo Caselles-Dupré, Pierre Fautrel, and Gauthier Vernier, produced a generative adversarial network portrait in 2018. Titled Edmond de Belamy, the work belonged to a series of generative images called La Famille de Belamy intended to resemble a fictional family tree. The name Belamy served as a French pun honoring Ian Goodfellow, the inventor of generative adversarial networks. Printed on canvas and set within a gilded frame, the portrait depicted a blurry man surrounded by white space.
The dominant colors in the generated composition were brown and beige.
2. Training Data and Code
The collective used portrait images and generative-adversarial-network software to produce its Belamy series. Its process drew on openly shared code by Robbie Barrat, whose earlier experiments had also generated portraits. The final work was printed on canvas and carried a mathematical expression instead of an ordinary handwritten name. Its production would become a separate subject of dispute after the auction announcement.
The source of the shared code attracted scrutiny alongside the print.
3. Christie's Auction Announcement
Auction house Christie's announced plans to feature the portrait in its Prints and Multiples sale in New York City. The auctioneers marketed the piece as the first artwork created using artificial intelligence to be featured in a major Christie's sale. Pre-auction estimates valued the print between seven thousand and ten thousand dollars. The announcement attracted substantial international media attention regarding the intersection of computer algorithms and the traditional art market.
The painting was hung near a bronze work by Roy Lichtenstein.
4. Bidding Process
The portrait was offered at Christie’s in New York on October 25, 2018. Bidding exceeded the published estimate of $7,000 to $10,000. A telephone bidder won the work, and the recorded sale price, including fees, was $432,500. The transaction concerned the framed physical print, while the code and generated images could be discussed independently of that object.
The final price was recorded with the auction fees included.
5. Collective's Stated Intentions
After the sale, members of Obvious discussed their plans to use the proceeds for further work. Attention also turned to the tools that had produced the portrait. The collective had not developed every part of that process from scratch. Earlier code and examples remained available online, allowing other artists to compare the project with its predecessors.
Robbie Barrat publicly questioned the credit attached to the project.
6. Attribution Controversy
Barrat drew attention to his earlier released code and images when the auction brought the portrait wider attention. Members of Obvious acknowledged using his work within their process. The discussion concerned which technical and artistic contributions should be credited, rather than whether the buyer had paid the auction price. The existing online material supplied a record of the earlier work.
The collective’s use of shared code became part of the public account.
7. Public Discontent
Obvious responded to reporting that described the group as inventing AI portraiture. Its members explained their use of existing techniques and software, while Barrat’s comments continued to circulate. The sale had already taken place, so those exchanges concerned attribution after the auction. They did not change the price recorded for the canvas at Christie’s.
The credit discussion continued after the sale was complete.
8. Documented Legacy
Christie’s sold Edmond de Belamy in October 2018 for $432,500 including fees, above the published estimate. The sale concerned a physical print produced by the Obvious collective using a generative-adversarial-network process. Questions over credit for earlier shared code were discussed separately from the auction result. The record of the sale established its price and maker attribution without settling every disagreement over how software contributors should be credited.
The auction price was recorded; the credit dispute remained separate.