The Image Is Not What It Appears
We often assume that an image is something simply there, a visible object presented to our eyes. Yet, as Hans Belting argues in An Anthropology of Images, an image is never identical to its material support. Images occupy a paradoxical position: they rely on a medium to appear, but they do not belong to that medium. They must be activated by the viewer’s imagination, which draws the image “out” of the opaque material and turns the medium into a transparent conduit. In Belting’s formulation, the image exists in a state of suspension between presence and absence, between embodiment and disembodiment, between materiality and mental projection. It is not the medium that ultimately “holds” the image, but the viewer who gives it perceptual life.
From Mediation to Substitution
This instability becomes newly consequential in the age of datafied bodies. Once the medium of the image becomes databases, model weights, and computational systems rather than celluloid or canvas, the separation between bodies and their images does not simply expand, it is structurally transformed. The “image” of a person may no longer require a person at all. A digital model can be generated, iterated, and deployed without ever having stood before a camera; a dataset can circulate long after the human it references has withdrawn, aged, or refused consent. In this environment, images no longer merely detach from bodies, they begin to replace them.In contemporary conversations, terms like multimedia and mass media appear so frequently that the word medium risks losing any real conceptual depth. If we want to use the term in a meaningful theoretical way, we have to clarify what we mean when we say it. McLuhan famously describes media as extensions of the human body, technological forms that recalibrate how we sense and navigate time and space, rather than neutral channels that deliver information. Art history, meanwhile, tends to define a medium either as an artistic category or the physical material an artwork is made from. But neither of these definitions fully captures what is at stake when we talk about images today. Across media theory, a recurring insight is that the medium operates as the technical and material condition that makes an image perceptible at all. A medium is not simply a conduit; it is the condition of the image’s visibility, the material and technological ground that turns a visual event into something perceptible. The medium, therefore, is neither external to the image nor subordinate to it; it is the ground through which the image becomes thinkable, legible, and real.
The Fashion Data Double
On the surface, an AI-generated fashion model image looks simple enough: a glossy figure posed against a studio backdrop, clothes hanging perfectly, skin without pores or fatigue. It could be a screenshot from any luxury campaign, until you realize there was never a body in front of the camera. No model booked, no lights adjusted to her height, no stylist pinning fabric to her spine. Instead, what stands in for “her” is a composite built out of scans, datasets, and models that can be rendered in endless variations without ever asking for rest or consent. Data & Society’s research on “fashion’s data doubles” names this shift: the fashion model is no longer only a person who works in front of a lens, but also a datafied proxy, a version of her body extracted, stored, and redeployed through computational systems. The report shows how models’ measurements, images, and movements become training material for virtual lookbooks and automated try-on tools, and how these digital stand-ins can appear in campaigns the model herself never participated in. Her “image” moves on without her.
Disturbing Belting’s Triangle
This transformation presses on a question that has been hovering around our course all semester, and that Hans Belting formulates sharply in his anthropology of images: what exactly is an image, and what kind of body does it require? For Belting, images are never just things “out there.” They exist in a triangle: body, medium, image. The body is the living site where images arise and are perceived; the medium is the material support that makes them visible; the image itself hovers somewhere between the two, inner and outer, psychic and material, always dependent on imagination to “lift” it from its medium. We do not simply control images; they occupy us, inhabit our memory, and help us make sense of the world. If we take Belting seriously, then AI fashion models are not just a technical novelty. They represent a disturbance in this triangle. Here, the “image” of a person is produced without a living body in front of the camera. The medium is no longer film or sensor but a computational system. Yet these images still land in human perception; they still cling to our ways of seeing bodies, beauty, gender, and race. The question, then, is not only what data doubles are doing to the labor conditions of models, as Data & Society carefully documents, but what they are doing to the very definition of images and embodiment
What Belting helps us see, then, is that images are never just “out there” in the world; they are always routed through living bodies that remember, fantasize, and perceive (Belting 2011). Yet this anthropological focus on perception also has a blind spot. It tells us a great deal about how images inhabit viewers, but less about how images extract from those who are pictured, or from those whose data underwrite the image in the first place. When the fashion model’s likeness becomes training data, what kind of “occupation” is taking place? The image does not just live in my memory; it also lives in a database owned by someone else. Belting’s triangle reminds us that images need bodies, but the data double forces us to ask a harder question: whichbodies, and on whose terms?
Black Boxes and Borrowed Authority
New media rarely establish credibility by announcing themselves as new. Instead, they tend to lean on the visual authority of older forms, adopting photographic conventions while quietly concealing their own mechanisms (Bolter and Grusin 1999).The AI fashion model is made to look as though it were photographed in a studio and the polished framing of editorial fashion imagery. The result is not merely an image that looks like a photograph; it is a medium that strategically disguises itself as the photographic, so that viewers inherit photography’s habits of belief without having to confront photography’s material preconditions. What emerges is a peculiar reversal of transparency. In Belting’s terms, the medium becomes “transparent” when the viewer’s imagination extracts the image from its support, treating the support as a conduit rather than an object (Belting 2011). With computational images, the conduit is transparent in a different sense: it is deliberately black-boxed. The viewer is invited to host the image while remaining structurally distant from the conditions of its production. This distance is not accidental; it is functional. It allows the medium to expand its role, from carrying to generating, without triggering immediate skepticism about presence.
Embodiment at the End of the Chain
Yet, if the initiating body is no longer required at the moment of production, embodiment does not disappear. It relocates. Hansen’s account of digital images is crucial here: digital mediation does not “free” images from bodies so much as it demands that bodies re-enter at the level of perception, affect, and sensorimotor completion (Hansen 2006). Even when an AI fashion model is generated without a photographed body, it still requires a perceiving body to be read as sensual, aspirational, racialized, gendered, desirable. The viewer’s body becomes the final site where the image comes to life, where the medium’s outputs are translated into lived sensation (Hansen 2006). The body is not erased, but displaced, arriving at the end of the process to authorize an image it never helped produce. This is where McLuhan’s idea of extension starts to feel uneasy. Media may extend perception, but they also change our sense of what attention should feel like, how fast it moves, how smoothly it flows, and what we come to expect from it (McLuhan 1964).Van Den Eede pushes the point further: every extension entails a diminishment, a redistribution of agency and awareness that often makes the extension feel “natural” precisely by making its costs difficult to perceive (Van Den Eede 2015). In computational imaging, the extension is not only optical; it is generative. The medium extends the image beyond the body’s presence, yet diminishes the body’s capacity to delimit, negotiate, or refuse what that image will become.
At this point, the “medium” is no longer simply the condition of visibility. It becomes an engine of substitution: a system that can model presence itself. And because it models presence in familiar photographic language, its substitution can be mistaken for continuity.If the medium can produce images without bodies, then the central political question becomes: whose bodies still pay the cost of visibility? The answer is rarely “no one.” Instead, bodies are translated into resources, and the image becomes a site where extraction can continue under the sign of realism. The Data & Society account of “fashion’s data doubles” names a structural reconfiguration: the fashion model is no longer only a worker who appears before a camera but also a datafied proxy whose measurable attributes proportions, facial geometry, movement, skin texture can be stored, and redeployed across campaigns and platforms (Data & Society). The decisive shift is not simply that images circulate; images have always circulated. The shift is that circulation can now occur with a reduced need for participation from the original subject. The model can be absent, asleep, unwilling, or contractually excluded, yet the proxy continues to “work.”
John Berger’s account of reproducibility helps frame this mobility historically. Once images become reproducible, they loosen from the singular contexts that once anchored them and acquire a new social life (Berger 1972). In AI fashion systems, that “social life” takes on an industrial form: the image is not only reproduced; it is *iterated.* It becomes parameterized, tweakable, and scalable. This is why the question “What happens to mediation when images can keep working without bodies?” is not metaphorical. It describes a literal labor shift: the work of appearing can be separated from the worker who once supplied appearance. The concept of the data double becomes most politically legible when read through Belting’s claim that images are neither identical to living bodies nor reducible to inanimate objects.
The deeper issue is that AI converts bodily work into a durable productive asset, a form of labor that can outlive the worker’s presence (Data & Society 2024). McLuhan’s extension thesis clarifies why this can feel strangely normal. Extensions do not announce themselves as domination; they present themselves as convenience, as “just how things work now” (McLuhan 1964). Van Den Eede adds the missing mechanism: when an extension becomes naturalized, its costs become harder to perceive; the body that made the extension possible is quietly erased from the story the medium tells about itself (Van Den Eede 2015). The data double is precisely this kind of naturalized extension: the model’s bodily labor is extended into a technical system, and the extension quickly becomes treated as the primary reality. The original body is reframed as merely the raw input.
This is why “endless labor” is not only metaphorical. It is structural. Tiziana Terranova’s argument about digital “free labor” helps explain how value extraction can persist without appearing as labor at all, because participation, capture, and circulation are built into the environment rather than enforced as discrete acts (Terranova 2000). Zuboff’s “surveillance capitalism” names an adjacent logic: systems thrive by turning lived experience into data that can be repurposed without reciprocal control (Zuboff 2019). Data doubles operate along this axis. What disappears is not work, but the conditions under which work can be recognized, negotiated, or refused.
A deeper anthropological question follows:
If images can be engineered to keep producing value after the body withdraws, what happens to refusal as a bodily capacity?
If image-making is understood anthropologically rather than purely technologically, then the current crisis of images appears less as a sudden rupture and more as a breakdown in correspondence.
From this perspective, the danger of AI-generated imagery is not that it fabricates images, but that it fabricates them without requiring continued bodily negotiation. Images no longer need to answer to fatigue, refusal, vulnerability, or time. They persist independently of the bodies that once grounded them. This is where image philosophy begins to slide, almost inevitably, into political economy. When images no longer negotiate with bodies, they become ideal vehicles for extraction. The data double exemplifies this shift. A likeness captured once can circulate endlessly, generating value without requiring further participation from the person it resembles. What is lost is not only labor compensation, but the body’s capacity to intervene in its own representation. The image no longer responds; it simply continues. In this sense, the problem is not that images misrepresent bodies, but that they no longer depend on them.
A Crisis of Relation
Anthropologically, this marks a profound transformation. Images have historically functioned as sites of exchange，between life and death, presence and absence, self and other. Funerary images, mirrors, shadows, paintings, and photographs all required the body to remain meaningful. Even when images abstracted or idealized, they retained a trace of bodily limitation. AI images, by contrast, risk becoming images without consequence. They do not age. They do not resist. They do not withdraw. And because they circulate with the visual authority of older media forms, they are often accepted without question.
Yet the issue is not simply deception. As Flusser warns, the true power of technical images lies not in their capacity to lie, but in their capacity to reorganize perception until their conditions of production disappear from view. Once images feel natural, their authority becomes difficult to contest. Over time, bodies that cannot match the smoothness, efficiency, and availability of synthetic images begin to appear excessive or insufficient by comparison. The image no longer reflects cultural values; it quietly installs them.
Seen this way, the crisis of AI imagery is not a crisis of realism, but a crisis of relation. When images stop corresponding with bodies, they cease to function as mediators and begin to operate as autonomous systems. They no longer translate human experience; they overwrite it. And because images structure how the world becomes intelligible, this autonomy carries real consequences，for labor, for aesthetics, for gender and racial politics, and for the very concept of embodiment.
The task, then, is not to abandon image-making nor to nostalgically recover a pre-digital past. Anthropology teaches that images are unavoidable. We live in them, think through them, and remember with them. The question is whether images can still be made to correspond—to materials, to bodies, to lived limits. Without this correspondence, images risk becoming a runaway cultural force: endlessly productive, endlessly circulating, and increasingly detached from the human conditions that once gave them meaning. In the end, the problem is not that images have gained power, but that they have lost negotiation. And once images no longer need to negotiate with bodies, it is no longer clear how bodies can negotiate back.
Works Cited
Belting, Hans. An Anthropology of Images. Princeton University Press, 2011.
Berger, John. Ways of Seeing. Penguin, 1972.
Bolter, Jay David, and Richard Grusin. Remediation: Understanding New Media. MIT Press, 1999.
Data & Society. Fashion’s Data Doubles. Data & Society Research Institute, 2024.
Flusser, Vilém. Towards a Philosophy of Photography. Reaktion Books, 1983.
Hansen, Mark B. N. Bodies in Code: Interfaces with Digital Media. Routledge, 2006.
McLuhan, Marshall. Understanding Media: The Extensions of Man. McGraw-Hill, 1964.
Terranova, Tiziana. “Free Labor: Producing Culture for the Digital Economy.” Social Text, vol. 18, no. 2, 2000, pp. 33–58.
Van Den Eede, Yoni. Tracing the Medium: Technological Mediation and Postphenomenology. Lexington Books, 2015.
Zuboff, Shoshana. The Age of Surveillance Capitalism. PublicAffairs, 2019.
