Thinking about images
Photographs, digital pictures, selfies, screenshots, memes, emojis, avatars, telescopes, microscopes, security cameras, images produced by machine vision, satellite images, radars, X-rays, sonograms, thermal images, media, cinema, series, film stills, scientific images, technological images, sketches, diagrams, maps, infographics, paintings, drawings, murals, graffiti, tattoos, stained glass, cave paintings, flags, banknotes, stains, shadows, silhouettes, fossils, footprints, fingerprints, reproductions, reflections, mirages, camouflage, mental images, imprints, imaginary images, hallucinations, apparitions, dreams, afterimages, memories, icons, symbols, fantasies, deepfakes, filters, glitches... These we find under the name of images.
Images are not just illustrations, but a way of thinking. This text deals with the ways in which we relate to them.
Why do we need to think about images?
There are now more images in the world than we can usefully count. It wasn’t always like this. In the past people might have had to make a pilgrimage to see an image: they walked through mountains and towns to get to the place where images were kept. Nowadays, the number of images you are able to see depends on the speed your thumb moves.
Just a couple of years ago, in 2021, numbers looked like this: YouTube viewers watched over one billion hours of video on the platform every single day, Instagram saw more than 300 million new photos uploaded daily. In one minute alone there were 510,000 comments posted and 293,000 statuses updated. Not only the figures changed, but also the vocabulary. The current figures are bigger and vaguer. Something like two trillion photographs were taken in 2025, according to market researchers.
We do not yet know how many images were generated rather than photographed. The companies publish their own totals: Google has said more than 50 billion images were made with its Nano Banana models by May 2026, and Adobe reported more than 24 billion Firefly assets by June 2025. The cost of an image now is somewhere not visible, in the electricity and water that train and run the models, in the data centres that hold what you upload. There is an illusion of free images as they get produced in quantities, responding to whatever we imagine we want to see. The cost of pilgrimage was different.
The increased rate of image production tracks the productive power of human cultural technologies. These began with the natural earth minerals and engraving tools of the Palaeolithic, passed through the politics of pigments in the medieval and Renaissance workshop, evolved through the printing press of the early modern era, took important expression in present-day methods of digital production, and now extend to systems that produce images without a camera and without a hand. This technological expansion means that human culture is predominantly image based, and that no appreciation of this culture is complete without a reflection on images.
Moreover, we do not just see images in the external world; we also think in images. Sigmund Freud declared that our unconscious is built in images, rather than words. Imagination? Eye effect? Visual perception? Saccadic ocular movement? All of these deal with the transmission of visual data between what is outside the skull and what is inside it. We encounter internal images in dreams, hallucinations, and at the borders of sleeping and waking.
In what ways can we classify images?
Researchers in visual studies have proposed many ways to classify images. One such schema is W. J. T. Mitchell’s genealogy of images (Mitchell 1984), which classifies images into five categories:
1. Graphic: Pictures, statues, designs, paintings
2. Optical: Projections, mirrors
3. Perceptual: Images collected by the senses
4. Mental: Dreams, imagination, creativity, ideas, memory, fantasmata
5. Verbal: Metaphors, descriptions, writing etc.
This is a fruitful conceptualisation of the world of images because it outlines the conflict under the name of “image”. But it is important to realise that models only approximate reality, there is no such thing as the perfect model of the image.
Notice what has happened to the fifth category. When Mitchell wrote, in 1984, the verbal image was the odd one out: the metaphor, the description, the picture you build out of a sentence. It is now the category the others come out of. A text-to-image model takes a verbal image and returns a graphic one. The prompt is the description, and the picture is what the description produces.
Mitchell’s classification shows that there are mental images as well as visual images. Image comes from imago, which means imitation and imagination. An image therefore has an appearance; it is seen. But an image can also be a mental construction, an imagination. Therefore an image is not solely a representation of something. The image has a reality in itself. In other words, images are not a mirroring of the world, an imitation of it, but images create the world, they produce meaning.
Images are never isolated or autonomous entities. Instead, they are deeply relational, dialogic, and linked to other images, texts, and media through networks of intertextuality, quotation, and intermediality (Bal 2002). Meaning is not produced in a single image, but in its relation to others. Every image belongs to a constellation. Aby Warburg’s Mnemosyne Atlas was a radical way to write art history, not as a linear process but as a field of recurrences. Between 1924 and 1929 he pinned photographs onto wooden boards covered in black cloth and by the end, around a thousand images across sixty-three panels. An antique relief could sit beside a Renaissance painting, a postage stamp or a newspaper photograph, because what interested him was not the period an image belonged to but the gestures that kept returning across periods: what he called the afterlife of antiquity.
This is why pictures can be destroyed, but not images. Images don’t disappear. Their material presence might not be here anymore, but they can survive in narrative. Take, for instance, representations of the Prophet Muhammad: these are images that wield power in virtue of not being visible. This is an example of an image surviving in a narrative when the image itself is absent.
Another example is in relation to resolution. When the resolution of visual images was coarser, there was more space for imagination. Viewers were able to interpolate into the image. But when the definition of the images grew, people needed less cognitive and perceptual effort to engage with the image. High definition images direct imagination and therefore leave less room for us to freight the image with our desires and expectations. Susan Sontag made a related argument, saying that what dulls our response to images is their number. “Images transfix. Images anesthetize”, she wrote. An event known through photographs becomes more real than it would otherwise have been, “but after repeated exposure to images it also becomes less real” (Sontag 1977, ch. 1). An image that asks less of you gets less from you. “Photographs shock insofar as they show something novel”, she writes, and then notes how quickly the novelty runs out.
On a current phone, the detail in a high-resolution image is often not recorded but inferred. In 2023 people noticed that Samsung phones produced remarkably crisp photographs of the moon, and argued about whether the craters had been photographed or supplied by the model that processed the picture. Google’s Best Take assembles a group portrait out of several frames, so that everyone is smiling at a moment that never happened. Upscaling software invents the detail it claims to restore.
The same logic has begun to work on faces. Filters on Instagram, Snapchat and TikTok do not so much decorate a face as propose a different one. Surveys by the American Academy of Facial Plastic and Reconstructive Surgery have tracked a rising share of surgeons reporting patients who want procedures in order to look better in their own photographs: 42 per cent of surgeons in 2015, 55 per cent in 2017, and higher again in surveys since (Rajanala, Maymone and Vashi 2018).
So the viewer’s imagination has been displaced by resolution, at the same time as it has been displaced by the camera’s representation. And representation has in turn, modified reality.
Can we define what an image is?
Though no definition can satisfy all images, definitions can be helpful in clarifying our intuitions about images. Vilém Flusser opens his philosophy of photography with one: “Images are significant surfaces” (Flusser 2000, 8). A surface, because an image is something you take in at once. Significant, because that surface carries meaning. We can extend this: an image is a significant surface that connects two or several points, and its structure is the assemblage of information and stories, the different elements that compose it. But rather than centring on the definition, we can centre on the meaning that images produce.
James Elkins, rather than deciding on a definition of what an image is, lists the theories that already exist: images as very thin skins of things, which is Lucretius; images as reminders of love; images as kisses; images as models. One of them says: “Images as reminders. This is, for instance, Susan Sontag’s position: images don’t tell us anything, they remind us what is important.” (Elkins 2011, 4)
Images have different meanings. Some images are more symbolic, more utilitarian. Compare, for instance, the kinds of images produced in art with the informational, schematic and notational images found in scientific visualisations. The expressive meaning of an image is what opens its symbolic interpretation. An image then ranges from being more informational and utilitarian, to more symbolic and expressive. And if we were to ask you to close your eyes and think of an image, it would probably be in the realm of the metaphorical.
The points that an image connects need not always be visible. This is where the symbolic lives: not in what an image shows, and not in what it withholds, but in the movement between the two. Compare the hyper-visible images of pornography with the more subtle images of erotica. Pornography gives direct representation to sexualised content, and in giving everything it leaves nothing to be interpreted. Erotica makes the absence of visibility salient: something is held back. Walter Benjamin put the general case about beauty: “the beautiful is neither the veil nor the veiled object but rather the object in its veil” (Benjamin 1996, 351). The image is neither what is shown nor what is hidden. It is the relation between them.
Vision: how do we see images?
Vision occurs in optical perception, where the visible spectrum is read through the organs of sight. Perception is visual, optical and ocular. Nonetheless, to see is a cognitive act as much as it is a physiological one. Cognition and vision interact in a reciprocally informative way in the act of perception. As John Berger said, “every image embodies a way of seeing”, and every person sees an image in their own unique way.
Images pass quickly without us noticing the majority of them. Vision is composed by the visual system, that contains the visual field, the optical lens, and the retinal and brain processing. Think of one person observing an object and another observing it having taken hallucinogenic substances. Our perception partly builds reality, but this reality is not exhausted in our perceptions. Another example is colour blindness, which is the deficiency to see colour or differences in colour.
But seeing is not uniform. Take for instance perception verbs like looking, paying attention, glimpsing, peeping, spying, noticing, observing, catching, comparing, picking out, detecting, having sight of, setting eyes on, surveilling, watching, keeping an eye on, shadowing, snooping. All these forms of looking take different shapes when engaging with an image. Sometimes they result in reflexivity, attentiveness or awareness.
MIT neuroscientists find the brain can identify images seen for as little as 13 milliseconds, but they’re still uncertain of how long the images can be retained in the brain. In order to see properly there has to be time: slowing down the perception, re-imagining what the picture depicts, and allowing for contemplation. Furthermore, understanding that the language of images does not equate to the language of text, and perhaps images can’t be fully read, but need also to be experienced.
Predictive coding models of cognition and perception argue that many of the images that we perceive are predictions: our brains make a best guess as to what the world contains, and then check if the prediction is true. This means we often see what we expect to see instead of what’s actually there. This leads to phenomena like change blindness, where people do not notice even substantial changes to their sensory environment.
Your brain predicts, and then checks. The phone predicts, for instance the moon, or a smile, and does not check. Both produce an image that is partly a guess. Only one of them is capable of being surprised.
Seeing with the other senses
Though sight is often privileged as the dominant sense, there is good evidence that all the senses interact when representing the world. On the one hand, linguistic data shows that even abstract concepts are represented using human sensory and motor schemas that draw on all the senses. On the other, neuroscientific evidence shows that sensory impressions are cross-coded: the brain can predict tactile sensation from a visual input, and vice versa. As the USC researchers put it, “Human brains capture and store physical sensations, and then replay them when prompted by viewing the corresponding visual image.”
Sight is processed in the visual cortex of the brain, as well as in the retina. But what about blind people? People who have been blind can repurpose the visual cortex for processing tactile data, encoding touch where sighted people encode images. This indicates that there may be processes common to all forms of perception.
Interactions between the visual and the tactile senses become more salient when the scale of viewing is increased. For example, Vermeer’s Girl With a Pearl Earring has been represented as a 10-billion pixel interactive image. Or Pluto’s true colours taken by NASA’s New Horizons spacecraft. From a distance, seeing is optical, but the closer you get, the more tactile it gets. “Haptic visuality” is Laura U. Marks’s term for this (Marks 2000): a way of looking in which the eye behaves less like an instrument of distance and more like an organ of touch, moving across a surface rather than through it towards an object. It is a visuality with tactile effects, triggering further memories like smell, touch and taste. We touch what we see with our eyes.
In terms of sound, sight and hearing cooperate in spatial awareness. The brain’s reasonable assumption that sensory information will cohere can lead to counterintuitive outcomes. See, for instance, the McGurk effect, where the incongruence of visual and auditory speech is not noticed and results in hearing something the voice is not saying. Another way of putting it is how music is represented visually.
The cross-mapping of visual and auditory stimuli is grounded in very fundamental aspects of human perception. Take the Kiki-Bouba effect: across every culture in which it has been tested, individuals associate the nonsense word “Bouba” with a round shape and the word “Kiki” with a jagged shape. This indicates that we cannot easily consider images without bringing in other sensory modalities.
The visual and the visible: how does our social and cultural context shape our reception of images?
Vision refers to the act of seeing, which is perceptual, but what about what is seen? Vision and visuality are not the same: the first is a physical activity and the second a socio-cultural construction.
Visuality and visibility are the complement to visual perception. These terms relate to our social and cultural relations with the visible world. Visuality has to do with lived experiences and social relations in the current technological ecosystem. Visuality therefore is the culturally and socially coded way in which we read the world. Looking and being looked at, knowing that the way we look, our gaze, is coded. Our gaze is modified by attentive desire or fear.
In terms of visibility, pictures are sometimes assigned “magical” powers. Some cultures are constructed around them (idolatry) and some cultures want to destroy them (iconoclasm). Because images exist, it doesn’t mean we can always see or access them. Social media platforms control what is visible. Images are constantly being censored, hidden, blurred. Part of what we see is mediated by the image’s visibility, what is represented and presented to us.
Images permeate and influence culture, and are therefore means for the support of power, or for resistance to it. Think for instance of the manipulation of pictures in false news, or the mainstream representation of women’s bodies. Add to this the invisibility of some subjects and cultures, set against those who hold the power of being represented. This is not only a question of who appears in images. It is a question of who held the camera, whose archives were kept, whose images were catalogued as evidence rather than as art, and which collections can be seen online and which cannot. Take for instance the Netflix documentary Disclosure, that outlines topics of transgender misrepresentation and invisibility.
Despite the inequality of visibility, contemporary culture is hyper visual. We are immersed and surrounded by images. Our daily lives are flooded by brand, commercial and self-imagery. Popular culture and spectacle is based on images and the resignification of them. Modes of control are also image-based: take for instance biometrics and surveillance. The human body no longer needs to travel to places to depict events. There are images produced by drones, microscopes, telescopes, security cameras and machine vision.
Many of these images are not made for human eyes at all. Harun Farocki coined the term operational images for them, in the Eye/Machine installations he began in 2000: images that “do not represent an object, but rather are part of an operation” (Farocki 2004, 17). They are made to be processed rather than looked at — to measure, to navigate, to track, to match a face against a list, to aim — and Farocki’s point was that they have no audience and no social purpose in the ordinary sense. They are not for edification, and not for reflection. Jussi Parikka has recently traced how far they have spread (Parikka 2023). An image that nobody ever looks at is still doing something.
Until recently, the striking example here was the Generative Adversarial Network, or GAN: an algorithm that could produce a photorealistic image of a person who does not exist.
GANs have since been overtaken by diffusion models, and the photograph of nobody has stopped being uncanny. It has become ordinary. The same systems now generate video, and the question has moved on. It is no longer whether you can tell that an image was generated. It is whether anything about the image lets you find out.
That question has produced an infrastructure of its own. Content Credentials, developed by the C2PA coalition, attach a record of an image’s history to the image itself, and platforms have begun labelling media they detect as AI-generated. This is a visibility problem rather than a technical one: metadata is routinely stripped when an image is uploaded and recompressed. The record of where an image came from is itself one of the things that can be made invisible.
The harm is not evenly distributed. Nudification apps and image-based sexual abuse overwhelmingly target women and girls. Fabricated images circulate around elections. And there is a question of labour. The models that generate these images learned from images other people made: photographs, illustrations, paintings, scraped at a scale no permission process could keep up with. Who is paid when an image is generated in the style of a living artist is now being argued in courts. It is also a question about images: whether an image can be learned from without being used.
Conclusion
Baudelaire stated that imagination is that faculty “which first perceives (...) the intimate and secret relations of things, correspondences and analogies, [so] that a wise man without imagination only seems a false wise man, or at least an incomplete wise man”. Our ability to put together and compare images gives rise to contradictions, similarities, assemblages, compositions and clustering. This is why images give us symbols, interpretations, arguments and sense.
In this text we talk about images as physical surfaces and as mental imaginations, and conclude that it is a concept that can’t be defined, but that this shouldn’t stop us talking about them. It leads us to think of vision as a physiological, as much as a cognitive process. In addition, we outline the way in which the senses cooperate, and how we can touch with our eyes or hear what an image is saying even if it’s not accurate. This allows us to engage with a visual world that is extended.
But vision is culturally and socially coded, which relates to visuality. Visuality is the intention we put in our gaze. We conclude with visibility, which is the amount of participation and representation something gets in our experience of culture and politics.
Jean-Luc Godard said somewhere that art deals not with the image of reality but with the reality of the image. This is why we should always ask ourselves: what is the story of the image?
References
Bal, Mieke. 2002. Travelling Concepts in the Humanities: A Rough Guide. Toronto: University of Toronto Press.
Benjamin, Walter. 1996. “Goethe’s Elective Affinities.” In Selected Writings, Volume 1: 1913–1926, edited by Marcus Bullock and Michael W. Jennings. Cambridge, MA: Harvard University Press.
Elkins, James. 2011. “Introduction.” In What Is an Image?, edited by James Elkins and Maja Naef. University Park: Pennsylvania State University Press.
Farocki, Harun. 2004. “Phantom Images.” Public 29: 12–22.
Flusser, Vilém. 2000. Towards a Philosophy of Photography. Translated by Anthony Mathews. London: Reaktion Books. Originally published 1983.
Marks, Laura U. 2000. The Skin of the Film: Intercultural Cinema, Embodiment, and the Senses. Durham, NC: Duke University Press.
Mitchell, W. J. T. 1984. “What Is an Image?” New Literary History 15 (3): 503–537.
Parikka, Jussi. 2023. Operational Images: From the Visual to the Invisual. Minneapolis: University of Minnesota Press.
Rajanala, Susruthi, Mayra B. C. Maymone, and Neelam A. Vashi. 2018. “Selfies—Living in the Era of Filtered Photographs.” JAMA Facial Plastic Surgery. doi:10.1001/jamafacial.2018.0486.
Sontag, Susan. 1977. On Photography. New York: Farrar, Straus and Giroux.
Sontag, Susan. 2003. Regarding the Pain of Others. New York: Picador.