Past as performance

This article is in collaboration with

The dog Argo recognises Odysseus even when disguised: what did he see? Artist Paolo Bufalini’s project, produced by the cultural organisation Sineglossa, bridges the latent space of dreams with that of informatics, where artificial intelligence generates images from data that are neither interpretable nor controllable. A space of the plausible, halfway between what happened and what could or will happen, which revisits the question of how much our recounting of the past is a performance. We met him.

Open 24/7: the mantra of contemporary capitalism is the ideal of a life without pauses, active at any time of day or night, in a sort of global continuous wakefulness, where the public intrudes on the private and sleep is forsaken¹. How much do we perform in our dreams? How do imagination and reality blend during sleep?

The plausibility of the dream state is at the heart of Argo, the exhibition by Paolo Bufalini, curated by the cultural organisation Sineglossa and created with the support of SIAE and the Ministry of Culture as part of the Per Chi Crea programme. It is on display until 5 December at the Palazzo Ducale in Genoa and from 12 to 15 December at the Home Movies – Archivio Nazionale del Film di Famiglia foundation in Bologna, as part of the The Next Real showcase. Eleven posters feature eleven images created using generative artificial intelligence, based on the digitisation of over 1,300 family photos.

Sponsored ad

Sponsorizza con noi

Paolo Bufalini (Rome, 1994) is a visual artist who lives mainly in Bologna. His work is characterised by cross-media approaches and marked formal heterogeneity, focusing primarily on exploring relationships between psychological, temporal, and material depths. His works have been exhibited in institutional and independent spaces in Italy and abroad, including the Gubbio Biennale, Dolomiti Contemporanee, Fabbri-Schenker Projects in London, Raum in Bologna, and Neverneverland in Amsterdam. Recent accolades and residencies include SIAE – Per Chi Crea (2023), the Emilia-Romagna Region Acquisition Award (2020), and the Nuovo Forno del Pane residency at MAMbo, Bologna (2020). (ph. Jacopo Belloni)

Why Argo and what is it?

Argo, from the title itself, represents the idea of a journey: a journey back in time and through the space of my home and family. It is the journey I took while revisiting family photos and the pivotal moments of my relatives’ lives. It is also Odysseus’ journey into the technological unconscious—latent space in informatics—where generative models create images based on data that are not directly interpretable.

The moment when Argo, Odysseus’ dog, recognises his master despite his different appearance seemed significant in relation to the generative AI system I chose to use. It is a system of mediation capable of recognising, but not “seeing”, the appearance of people in photographs, following an order of knowledge different from that of humans.

The images on display are entirely synthetic because this project is also a journey into illusion: my father, mother, and sister are portrayed as something that never was and will not be but could have been.

Ultimately, the exhibition shows them to you but does not tell their story: after seeing it, you will still know nothing about them. This is because the images are plausible but not real, and they condense imagination and factuality, as dreams do.

What process led to the generation of the images?

I scanned around 1,300 family photos spanning from the 1950s to the early 2000s. The first step was sorting and classifying the images to identify those potentially usable, focusing on ones with clearly recognisable subjects. Then, I trained pre-existing text-to-image generative models to plausibly reproduce the likenesses of my parents and sister, based on my textual requests, and to contextualise them within new images. I used between ten and thirty photos for each model and trained about twenty models in total—at least six or seven for each subject.

For the actual generation process, I used a node-based interface², where I defined the generative parameters and the characteristics of what I wanted to represent—for example, an image of my father—specifying detailed attributes such as clothing, pose, lighting, and composition. In addition to textual requests, or prompts, I adjusted various parameters, including, for instance: the level of similarity required between the generated image and the request, the number of iterations for processing³, and the data sampling method. It was a very methodical and empirical process, where I slightly altered one parameter for each new image generation to observe the outcome.

For example, to create an image depicting my father, my prompts always followed a specific order:

“MBY” (Marcello Bufalini Young, the subject’s unique identifier), “man” (reference category), and “eyes closed” were the first three elements, followed by descriptions of the setting: type of clothing, hairstyle, lighting, shot composition, and occasionally references to cinematic or photographic film names to guide the image’s aesthetic.

The images I obtained often deliberately resemble cinematic stills or analogue photographs. Furthermore, because it was impossible to perfectly clean the scans for every input, dust specks, scratches, or stains on the photos were incorporated into the model’s training, as if they were inherent features of the depicted subjects. For this reason, the resulting images have a distinctly analogue feel.

Why did you choose to depict only sleeping people?

I have long been interested in dreams as a space where different fragments of reality merge: a form of condensation akin to the process I used to create images from biographical material, blended with the reworking of generative AI and the visual paradigms already embedded within the models I modified. Sleep is a moment when a person is simultaneously present and absent, suspended in a reality that frees them from performance. Even before Argo, in 2020, my installation Proposal displayed two cushions moving in synchrony, simulating a person’s steady, slow, and silent breathing that remained imprinted on the objects.

The development of that work happened to align with my reading of 24/7 (Einaudi, 2015) by art critic Jonathan Crary: a book discussing how, in our contemporary technological and capitalist world, sleep is increasingly eroded—starting with the excessive light from phone screens, streetlights, and televisions. We are expected to stay awake because during sleep, we do not perform, produce, or consume. Although this aspect is not directly tied to Argo, I find it compelling to think of sleep as a space where, precisely because of its minimal degree of performativity, identity dissolves and becomes more malleable.

I decided to investigate sleep as both a moment of absence or reduction of the performance that we experience when we are awake, and as a moment of vulnerability, intimacy, and privacy, always connecting it with “the machine”, or technology.

Proposal itself holds these two elements together: the intimacy of bodily breathing and the synthetic way it was staged.

In this sense, in Argo, the decision to depict all subjects as sleeping exempts them from their identities, transforming them into pure images and tending towards the dematerialisation and impersonality of icons. For the prints, I chose a medium-grain cotton paper that makes them softer and slightly textured, aligning with the images’ inherent blurriness and the extremely smooth transitions between different parts of the image—such as the pillow and the face—which sometimes seem to blend into each other, veering towards indeterminacy and abstraction.

At the centre of the exhibition is a sculptural piece with vials. What do they represent?

The two vials contain fluorescent green liquid swirling inside thanks to two electromagnetic field generators. The liquid is composed of second-hand jewellery—two earrings and a gold ring—dissolved in an acidic solution I created in collaboration with the Chemistry Department of the University of Turin, under the supervision of researcher Andrea Jouve.

The process is chemical but also recalls alchemical principles of regeneration and infinite potentiality: it would be possible to re-extract physical gold from the fluorescent green solution, though not in its original form and with the same ambiguity as Argo’s images, based on real data but no longer true in their form.

I chose gold for its ability to “regenerate” and encompass various ideas of value—alchemical, emotional, economic, symbolic, even electric. I was inspired by the famous historical anecdote³ of Nobel laureates Max von Laue and James Franck, who protected their medals from Nazi confiscation by dissolving them in aqua regia and later reforging them. If you think about it, gold jewellery has always been a family heirloom passed down through generations, again touching on the question of identity. On the other hand, the graphics cards used in generative AI GPUs are often made of gold due to its high conductivity. In a way, gold is part of the material context that participates in the genesis of the images themselves.
 
I also liked the idea of having something amidst the images of sleeping figures that could symbolise an active process, interpretable in multiple ways: the dream process of those who are sleeping or the generative process of the images themselves.

One final curiosity about the exhibition?

In the images of Argo, two radically different elements intertwine: the private dimension of sleep and the visual codes of family photos. These are snapshots where performative elements—posing, smiling—alternate with spontaneity and innocence, typical of amateur photography. Clearly, fewer photos were taken in the past than today, and people were often portrayed at their best during celebrations such as birthdays, baptisms, or dinners. This almost omnipresent element of smiling then carried over into the generation of the images, where expressions are always cheerful, giving this sleep an air of serenity—almost a slyness.

 

1. See also Cray, J. (2015), 24/7, Einaudi ↩︎
2. A node-based interface is a software framework that operates on the principles of graph nodes. It is a type of algorithm that allows for the description of relationships between elements in a chained and linear manner, without feedback loops or retroactive connections. ↩︎
3. The processing of each image took between 15 and 40 steps. ↩︎
4. The German physicists Max von Laue and James Franck, both winners of the Nobel Prize in Physics in 1914 and 1925 respectively, sent their 23-carat gold medals to Niels Bohr’s Institute of Theoretical Physics in Copenhagen, to protect them from Nazi plunder during the Second World War. George de Hevesy, with the knowledge that the Nazis were on the streets of Copenhagen, dissolved the medals in turpentine. The resulting solution was placed on a shelf in the laboratory of Niels Bohr’s Institute of Physics. Later, when the war was over, de Hevesy precipitated the gold from the acid mixture and sent it to the Nobel Society, which succeeded in recasting it and reviving the two medals presented to Max von Laue and James Franck in 1952. For more: A unique gold medal. (n.d.). NobelPrize.org. https://www.nobelprize.org/prizes/about/the-nobel-medals-and-the-medal-for-the-prize-in-economic-sciences/ ↩︎