Generative Photorealistic Images as Nodes of Memory: Artistic Practices and Multisensory Archives in the Digital Era

The rapid development of deep machine learning technologies has led to the emergence of photorealistic images created using neural network algorithms (e.g., Midjourney 6). While researchers typically explore these images in contexts such as deepfakes, misinformation, cybersecurity, and the erosion of trust in visual historical evidence, they rarely examine how such technologies shape memory within everyday experiences of individuals. This article addresses this gap by analyzing artistic projects that employ generative photorealistic imagery as tools for structuring personal memories and narratives, thereby transforming them into nodes within a heterogeneous network of inclusive, non-institutionalized collective memory. Drawing on niche construction theory (Heersmink, 2020) and sensory ethnography (Järviluoma and Murray, 2023), the article argues that these artistic practices are not intended to manipulate historical facts but rather integrate diverse personal experiences, resulting in multilayered, multisensory narratives that redefine collective memory beyond traditional audiovisual or text-centered frameworks.IntroductionIn recent years, we have increasingly encountered images created using neural networks that astonishingly mimic authentic photographic images with remarkable precision and detail. Some are actual photographs processed by deep machine learning algorithms, but others are entirely artificially generated frames depicting events that never occurred in reality. Russian researchers of contemporary photography Fadeeva, Persheeva, and Pronina note that ”image generation technologies capable of creating impeccably realistic images are taking us further away from realism, from the indexical nature of photography, and its value as evidence.” (Fadeeva et al., 2024) This distancing in contemporary digital art has become the foundation for an entire movement at the intersection of photography and generative graphics using graphic neural networks. Researchers worldwide have noted that ”photo simulations created by neural networks hold an intriguing status – they do not depict reality, yet at the same time, they illustrate it.” (Bylieva, 2024) Therefore, ”there is no direct relationship between the generated images and the status of a document, as they are an abstract representation of reality. However, there is an intrinsic relationship with this mass of information extracted from the real, appropriated, homogenized, flattened, abstracted, reconfigured, and transformed into a representation of reality that, despite not belonging to the realm of the real, has a significant ability to influence how we perceive the world around us due to its resemblance to photography.” (Luz, 2024)Various sources employ a wide range of metaphors to describe this artistic practice. For example, Manovich reflects on the so-called ”memory machines,” which ”extract information from existing cultural material during the training process, constructing a new historical archive.” (Manovich, 2023) In his experiments with image generation tools based on large language models (such as Midjourney), he creates what he calls ”historical fictions” – algorithm-generated images that mimic scenes from his past life in Moscow. These images resemble the author’s hazy memories rather than accurate historical photographs: a sort of ”remix” of how real people might have looked at that time and place. (Manovich, 2022)On the other hand, Fortunatov discusses the concept of ”neurohistoricisms,” meaning ”the empty symbolization of historical foundations in AI-generated images […] presented as depersonalized visual texts devoid of both a human presence and dramatic conflicts.” (Fortunatov, 2023) The resulting portraits offer a different perspective on historical material: the generative algorithms of large language models average out the features of real people, creating a single “face” in which individuality dissolves. At the same time, such a photograph loses what Susan Sontag described as evidence of existence. A photograph records a moment in reality, endowing it with uniqueness, authenticity, and emotional significance. It becomes a witness not only to what happened, but also to the individual view of it. It carries a trace of presence: that of the person portrayed, the photographer who took the picture, and the viewer who later looks at it. Neural network–generated images may seem “plausible,” but they erase personal uniqueness and turn history into an anonymous, glossy construct. This is not merely a new form of visual art – it is a change in the very nature of how we document and perceive the past. (Sontag, 2003)Nevertheless, Manovich does not entirely reject these characteristics of neural network-generated images. For example, in his project ”The Unpredictable Past” (2021), he generated group photographs of tenth-grade students from a Russian secondary school from 1966 to 2016. These portraits serve as an example of how an algorithmic ”memory machine” transforms many specific historical images into a generalized, averaged version, showcasing both the potential and the limitations of generative algorithms when working with visual data and historical or cultural heritage. Using generative algorithms, Manovich demonstrates how the past can be recreated, yet its unique details and nuances can be lost. At the same time, the author becomes a witnessing subject, whose memory confirm the authenticity of what is depicted.For instance in artist Yu. Kuznetsov’s 2021 work ”Archive of Paramnesia,” a neural network, trained on photographs from the artist’s own archive, generates images that hover between authenticity and forgery, prompting viewers to reflect on such contemporary phenomena as deepfakes and post-truth. Similar works include the photographs of a fictional earthquake in the United States, The 2001 Great Cascadia 9.1 Earthquake & Tsunami – Pacific Coast of US/Canada, which circulated on Reddit in 2023, or art projects like ”Anemoia” (2024) and ”The Family That Does Not Exist” (2023) by artist Victoria Gurova (Milagrelia). The latter is “a photo album” entirely composed of neural network-generated images depicting a typical family story—with the key difference being that none of the people, events, or places in the album ever existed. German artist Boris Eldagsen’s art project, ”Pseudomnesia,” combines his visions of abstract art from the 1940s with AI capabilities to write his own history of modern art. This history never happened, and no one was ever photographed.

Récupéré en direct depuis OpenAlex et désinversé. Les résumés ne sont pas conservés dans cette base de données : les index inversés représentent 8,6 Go des 9,3 Go de texte de la base, et le serveur dispose de 13 Go libres.