“It has been the special genius of our century to investigate things in relation to their context, to come to see the context as formative on the thing, and, finally, to see the context as the thing itself”1
Through animation, cardboard cutouts, and AI-generated memories, human + looking + art mines the familiar online imagery of people looking at art. Three interconnected works dissect the representation and consumption of online and offline art: The animated film art by the yard, reworks the scale of people admiring art on auction websites. conTemplate turns these same images into cardboard cutouts to populate otherwise empty gallery spaces. openings is an ongoing project about all the things I can remember from art openings, generated with text-to-image AI and probing the bias in models and navigating how to not fall foul of corporate AI rules.
In 1976, Brian O’ Doherty published three articles in Artforum examining the turn towards context in twentieth century art. These formed the basis for Inside the White Cube – The Ideology of the Gallery Space (Brian O’Doherty, 1986). In it, he investigated what the context of the precisely staged art gallery does to the art object, what it does to the viewing subject, and how the context actually devours the object, becoming it. The careful staging cuts off the outside world, a white blank world without windows, the light usually coming in from above, giving a sense of untouched timelessness and hushed focus. For Thomas McEvilley, this unmooring from a present, beyond time “implies a claim that the work already belongs to posterity – that is, it is an assurance of good investment.”2
Now, in the early 21st Century, in addition to the bricks and mortar gallery space, the virtual presence of artworks circulating in various auctions take place. Some of these auctions are the virtual twin of auction houses such as Sotheby’s and Christie’s, designed as online brochures for the sale of lots and also to make their brand relevant and visible to potential traffic.
In order for these ‘traditional’ auctions to keep their market share, they engage in online algorithmic auctions such as Google Ads, designed to determine “which ads should show with a lightning-fast ad auction, that takes place every time someone searches on Google or visits a site that shows ads.”3
When you browse the online catalogs of auction houses like Christie’s and Sotheby’s, you’ll often see pictures of people admiring art. Each of these pictures are made up of three key elements:
1. The artwork for sale – this takes centre stage
2. A person – they’re shown looking at the artwork
3. The setting – a backdrop that frames the scene
Whilst each of these elements varies slightly between each lot, they are a composite of three photographs, algorithmic collages, determined by a script that plugs into metadata that calculates scale. The art takes center stage on your screen, while the person and the setting are sized around the artwork to make everything look proportional, as if the art is being appreciated in an actual room. The same people and settings might pop up across different art pieces, striking various poses and positions, all to help you, the potential buyer, visualize and imagine making that artwork a part of your life.
conTemplate (cardboard figures plinths etc) and Art By The Yard (film)
In human + looking + art, these images of people become the entire focus of an exhibition, materialised as lifesize cardboard cutouts, populating a gallery space, and placed so as to look at an empty wall or plinth, or the occasional animated video of themselves projected onto a wall.
The animated film Art By The Yard takes screengrabs from the auction sites but with the images rescaled so that the humans are consistently the same size, shifting the focus onto the humans looking at art.
A 19 minute version slows the action down to 19 seconds per image, the purported average length of time gallery visitors look at an artwork.
The film takes its name from work made by Giuseppe Pinot Gallizio, co-founder of Situationist international, in 1958 where he made and sold art by the yard.
What cynics have long predicted suddenly came to pass: abstract art was on sale not by the painting but by the yard. In Munich’s fashionable van de Loo gallery, Italian painter Pinot Gallizio, 57, did a booming business by snipping his 10-and 20-yard canvases into appropriate lengths. Customers were free to choose according to their needs and pocketbooks“ 4
More than half a century later, and within a different situation, perhaps these auction sites are performing algorithmic reenactments of Gazzilio’s work, although collectively they are in the market for the highest price possible.
“Painting as such has reached the end of its road. From now on, the human eye will be perfectly satisfied by seeing any colour or shape, provided the colour is brilliant and the shape imaginative. You won’t get modern art any cheaper and certainly not any better“5
Computers looking at humans looking at art
The figures I used in conTemplate and Art by The Yard come from a typology of stock images that are, like in the Christie’s and Sotheby’s examples, use a content delivery system to make them appear as if actually in a scene. These, together with countless images of people looking at art in white cube galleries (some set up and photographed for real life), exist on public facing auction websites, in gallery archives and circulate within image search engine results that are designed to generate ad revenue and what is termed by Alphabet, Google’s parent company, conversion.
These same images are also scraped by researchers and corporations for use in designing text to image machine learning models. I searched for pictures of people viewing art using various keywords using Google’s advance image search function. I noted which people appeared, their clothing, and poses and created a database to anaylse with commercial computer vision labelling to see what patterns this might also reveal.
As I was collecting images and their labels, a new machine learning text to image preview feature – Firefly by Adobe – was released in early 2024. This release is trained on Adobe’s own stock images. I used their ecosystem to search for examples of stock images of humans looking at art. As with previous projects I’ve been experimenting with that examine corporate AI implementations of text to image software, I decided to use a stock image and generate a scene from the same stock image, at first repeating variations of the instruction people looking at art and in-painting areas of the image with the same instruction over and over. An example of the output of this is below.
This process made me reflect on the context in which these images both come from but are made. I decided to pull back the frame and show the construction process and the interface as an intrinsic part of the things I ma making, to give more of the context. This has resulted in an ongoing series of films called Openings, where I am making screen recordings of the entire interaction and interface – phrases inputted, waiting time, mouse cursor moving to erase an area to be replaced with the next command. I have also begun adding sound using text to sound sound effect models, which are much more rudimentary at the current stage.
Kypros Kyprianou, 2024
The Openings films each begin with the text string human+looking+at+art to generate images, but after a a few repeats of this, I begin to modify the text string slowly, eventually creating a narrative made from half remembered people and situations at art openings. Like the image above, these begin as a backdrop of a stock image of an empty gallery which I modify with text to image in-painted into specific areas, with several passes, pulling back as the scene grows as in the tableaux above. The interface used in generating the image is retained in the film. Below are some of the text strings from one series.
As can be easily surmised, Adobe’s text to image AI is plagued with similar problems of bias as all other generative AI systems I have so far encountered.