Lacan's Gaze and the AI Portrait

Who Is Looking at Whom?

"In the scopic field, the gaze is outside, I am looked at, that is to say, I am a picture."
— Jacques Lacan, The Four Fundamental Concepts of Psychoanalysis¹

I. A Portrait That Cannot Look Back

I keep coming back to a particular kind of image. It is a fully synthetic portrait generated by a text-to-image model — a face that has never had a body, eyes that have never opened, a person who does not exist. The lighting is precise. The pores are convincing. The expression is, in the language of the prompt I used, "pensive, just before turning away." Everything is there. And yet, when I sit with the picture for a while, I feel an odd cold settle in. The face is looking toward the camera. It is not looking at me.

This is not a complaint about realism. The realism is not the problem. The problem is closer to what an old photographer once said about an unsuccessful portrait sitting: that the sitter "had not yet come into the room." There is a difference between a face that is rendered with optical accuracy and a face that returns the look. The difference is small. It is also, in a strict sense, the entire ethical horizon of portraiture.

Jacques Lacan devoted four sessions of his eleventh seminar in 1964 to this problem, later published under the heading "Of the Gaze as Objet Petit a."² He called it the gaze. He did not mean the look of the spectator at the painting. He meant something more disturbing: the look of the painting at the spectator. The gaze, in Lacan's revision, is not what I direct at the image. It is what the image directs at me, from a point I cannot occupy, in a register where I am the one being seen. In this essay I want to take that reversal seriously and ask what it means in front of a portrait that was never seen by anyone before it appeared on my screen.

II. The Reversal: How Lacan Turned the Gaze Around

In ordinary speech we say that we cast a gaze. A man on a train casts his gaze across a field. A photographer casts her gaze across a sitter. The grammar is one of subject and object: I am the one who sees, the world is what is seen. Lacan, in Seminar XI, turned this grammar inside out. The gaze does not belong to me. It belongs to the field. I do not see from a sovereign point outside the picture. I see from inside a field that is already constructed, already lit, already arranged for a look. And that look — the look that arranges the field — is what Lacan calls le regard. The English translation gave us "the gaze," and the slight clinical strangeness of the word is a translator's gift, because in everyday English we do not usually say it about ourselves.³

The structural claim is simple to state and difficult to inhabit. There are two terms in the visual field — the eye and the gaze — and they are not on the same side. The eye is what I have. The gaze is what looks at me. I encounter the gaze whenever I have the sudden sensation, in front of an image or a scene, that I am not its master. The anecdote Lacan tells in the seminar is well known. As a young man on a fishing boat off the coast of Brittany, he was shown a sardine can floating on the water by a fisherman named Petit-Jean, who said, "You see that can? Do you see it? Well, it doesn't see you!" Lacan reports that he did not laugh as the fisherman expected him to, because he sensed at once that the joke was, in fact, profoundly wrong. The can was looking at him — not with a literal optic but with a point of luminous reflection that picked him out as one element in a scene he had not designed.⁴ He was not the photographer of that scene. He was inside its frame.

Joan Copjec, in Read My Desire, has insisted that this Lacanian reversal cannot be domesticated into a cultural studies argument about "who has the right to look." The gaze, for Lacan, is not simply a privilege held by one social group over another. It is a structural feature of the visual field as such — the way the field is organized around a point from which I am observed and which I can never reach.⁵ Bruce Fink, in The Lacanian Subject, helps clarify why this matters clinically and philosophically: the subject is not formed first and then placed before the image; the subject is produced within a field already structured by the Other's desire.⁶ Hal Foster, in The Return of the Real, takes the reversal into the history of postwar art, where the most demanding contemporary work confronts the viewer not merely with an image to consume but with a visual situation that exposes the viewer's own position.⁷ Across these readings the basic claim is constant. The gaze is not where you stand. It is where the picture has already arranged you to stand.

This matters in front of an AI portrait for a reason I will try to state carefully. Photography, since its invention, has carried within itself the trace of a gaze that was once active — the gaze of the sitter who saw the photographer, the gaze of the photographer who saw the sitter, and the gaze that comes back to me through the print as a residue of that mutual seeing. Roland Barthes, in Camera Lucida, calls this residue that-has-been: the simple, unsettling fact that the body in the picture stood in front of a lens, in this light, on this day.⁸ It does not matter whether the photograph is technically good. It matters that someone was there. And because someone was there, the gaze in the image is, however faintly, a returned gaze. The portrait looks back because there was a face that already looked.

The fully synthetic AI portrait removes that condition. The face in the picture was never anywhere. No one was photographed, because there was no photographed person. The composition is assembled from the statistical pull of other portraits and image conventions, and what comes back is a face-shaped surface that mimics the conventions of returning a look without ever having returned one. This is not a moral indictment of the medium. It is a structural fact about it. And it suggests that if AI portraiture is going to develop a serious visual ethics — and I believe it can — it will have to invent, not borrow, the conditions under which an image can be said to look back.

III. The Skull at the Bottom of the Frame

The image Lacan returns to in Seminar XI is Hans Holbein the Younger's The Ambassadors, painted in 1533 and held today by the National Gallery, London.⁹ Two ambassadors stand on either side of a table strewn with instruments — a celestial globe, a terrestrial globe, a lute, a torquetum, books — the entire equipment of Renaissance worldly mastery. The composition is upright, symmetrical, confident. The men are exactly the kind of subjects Renaissance perspective was invented to flatter. They have a place. They occupy it.

And then, at the bottom of the picture, there is a stain. A long, smeared, oblong thing appears between the men, tilted across the floor. If you stand in front of the painting and move toward a sharply oblique position near the lower right of the canvas, the stain shifts as you move. At a certain point the smear resolves. It is a skull. It has been painted in anamorphic projection, so that the position from which it appears in proper form is not the position from which the ambassadors, the torquetum, and the globes can be comfortably seen. The painting has two visual systems built into it, and they cannot be inhabited simultaneously. From the place where you see the ambassadors, the skull is a smear. From the place where you see the skull, the ambassadors are displaced.

Lacan reads this constructive choice as a parable of the gaze. The skull is not a decorative curiosity. It is the point at which the painting interrupts the fantasy of frontal mastery. It belongs to the work's memento mori logic, but it does its work by becoming illegible from the principal viewing position.¹⁰ The Renaissance perspective machine flatters you with a place from which the world can be measured and possessed. But the picture has installed a second perspective whose function is to remind you that this measurement is bought at a price — the price of mortality, of the skull, of the vanitas tradition that The Ambassadors knows it stands inside. You can be the master of the picture. Or you can see what is looking at you. You cannot do both without moving.

I keep this image in mind when I look at AI portraiture, because it gives me a working test. A good portrait, whatever the medium, has two perspectives built into it. There is the place from which the sitter is rendered — the careful, lit, composed position the viewer initially occupies. And there is a second place from which something in the image returns the look, often from an oblique angle. The second place is what makes the portrait live. It is what gives the eye somewhere to slip and catch and stay. It is the difference between a face and a face.

This is harder than it sounds to engineer in a generative system. The default behavior of many image models is to produce a coherent and immediately legible perspective, especially when asked for a "portrait." Public image cultures are saturated with frontal or near-frontal faces, and prompt-to-image systems often draw on those conventions with remarkable efficiency. When you ask for "a portrait," what you often get is the average of well-composed portrait conventions: the familiar orientation, the familiar lighting, the familiar distance. The smear at the bottom of the frame — Holbein's skull — is not in the average. It was a structural exception in 1533, and it remains a structural exception now. To put it back into an AI portrait requires that the prompt itself install the obliquity. We will come to that practice in the final section.

IV. Objet petit a and the Eldagsen Refusal

The third Lacanian term I want to bring to the AI portrait is objet petit a — what Lacan calls the object-cause of desire, the elusive remnant in a visual field that pulls the gaze and never delivers itself. Objet petit a is not an object that is simply missing. It is the structural fact that something is always missing, and that the looking is organized around what is not there.¹¹ A portrait, in this register, is not merely a record of a face. It is an arrangement around a cause that cannot quite be located in the face but which is the reason the face holds us. The catch-light in the eye. The tension between the brow and the mouth. The angle at which the head has just turned. None of these is objet petit a by itself. Objet a is the place in the image from which our looking is being conducted, and which the looking can never reach.

I think about this when I look at Boris Eldagsen's PSEUDOMNESIA | The Electrician, the AI-generated or AI-assisted image made in 2022 that was selected as a winner in the Sony World Photography Awards 2023 Open Competition, Creative Category, before Eldagsen refused the prize at the London ceremony on 13 April 2023.¹² The image is haunting. It is also, on a Lacanian reading, instructive. Eldagsen produced an image that performs, very convincingly, the conventions of mid-twentieth-century European studio portraiture — the soft side-light, the older woman behind the younger figure, the slightly faded greyscale that signals a recovered family archive. The conventions are doing their work. The viewer scans the image as if it were a remembered photograph, with the affect that attaches to that-has-been. And then — this is where the refusal comes in — Eldagsen disclosed the essential point: no one had been photographed. The image has no that-has-been. The conventions were running on absence.

It would be too simple to say that the image is therefore merely fraudulent. Eldagsen's own framing made the more interesting claim: PSEUDOMNESIA names a false memory, an image of a past that did not occur, generated through a process that has access to large archives of photographic convention and can recombine them into surfaces that feel personally remembered. The image is precisely an objet a of cultural memory. It is the object around which a viewer's looking organizes itself as if the viewer were remembering, when in fact the viewer has been positioned, by the image, in the place where remembering would happen if there were anything to remember.

I read Eldagsen's refusal as ethically clarifying for AI portraiture in general. In this reading, the refusal was not merely a confession of inauthenticity; it clarified that the image belonged to a new genus of visual artifact whose relation to photographic seeing is not equivalence but inversion. Eldagsen later adopted and popularized the term promptography for this genus, following Christian Vinces's suggestion, while insisting that the result is "not a photograph" but an image that uses "the photographic" as a visual language.¹³ The AI portrait does not record a gaze. It models the cultural surface that returned gazes have left, and offers that surface in place of a returned gaze. The viewer, encountering the surface, may feel the structure of being seen without being seen by anyone. This is close to the structure of objet a: the cause of looking that is not a thing one looks at.

For practitioners, the lesson is not to give up on AI portraiture but to take responsibility for what kind of objet a the image is constructing. A prompt that simply chases technical realism — "ultra-realistic, 85mm, soft window light, professional studio portrait" — tends to produce a smooth, average surface that returns no gaze in particular. A prompt that designs an obliquity — a turn of the head away from the dominant axis, a refusal of eye contact, a piece of the body cut off by the frame at exactly the place a conscientious photographer would have respected — begins to construct an image around a missing point. It will not give back Tarkovsky's candle. But it will at least admit, structurally, that someone is supposed to be looking, and that the looking does not arrive.

V. The Mirror Stage in Latent Space

There is one more Lacanian term that I cannot avoid in front of these images, and that is the mirror stage. In one of his earliest important papers, gathered into the Écrits, Lacan described the moment in which the small child, sometime between six and eighteen months, recognizes its reflection in a mirror and assumes the unified image as itself.¹⁴ The recognition is, Lacan insisted, a misrecognition. The child cannot yet coordinate its motor system. It does not yet possess, in lived terms, the unity that the mirror offers it. What the child assumes is an ideal image that comes from the outside, and the assumption is the founding act of the ego.

Most readers know the next move: the ego is, for Lacan, an alienation. It is not the deep truth of the subject. It is a defensive imaginary unity assembled around a borrowed image. The misrecognition is not corrected over time. It is layered, generation after generation, by other borrowed images that perform the same function — photographs, school portraits, passport photographs, professional headshots, social media avatars. Each time, the subject identifies with an image that is more whole than the subject feels. Each time, the cost is a small alienation. Each time, the structure repeats.

The generative image, I want to suggest, is the latest and strangest layer of this structure. When I generate a face that does not exist, and the face is recognizable as the kind of face a photograph could have made, I am participating in a mirror stage that has moved one level back. The mirror is no longer only the polished surface in front of which a child stands. The mirror is also the latent space of a model trained on vast image-text datasets drawn from online photographic culture, returning to me a face that is the statistical condensation of countless visible conventions.¹⁵ The face is not anyone's. But it is recognizable as a face. And in the act of recognition, the structure of the mirror stage repeats — with the difference that the image now has no body behind it and no body has, ever, occupied that place.

This is not a science fiction observation. It is a description of what is happening on screens now, at enormous scale. People generate faces that resemble them, faces that flatter them, faces that they could not have been, and the generation can function as a new form of self-image. The Imaginary register is being industrialized. And — to return the argument to the gaze — what is industrialized is precisely the side of the visual field where the eye sits. The model produces eyes by the thousand. What it does not produce by default, at least in Lacan's strict sense, is the gaze. There is no singular point in latent space from which the picture is looking back. There is a vast distribution of eyes that have looked, recombined into a surface that does not look from any single place.

For the AI artist who is also a photographer, this is a chance, not a loss. Because the gaze is missing as a default, the gaze can be made present as a choice. Every decision that resists the smooth average — every turn of the head, every dropped eye, every figure seen from behind in the manner of Caspar David Friedrich's Wanderer above the Sea of Fog, every frame in which the sitter is not where the conventions of the studio say the sitter should be — is a way of putting the gaze back into the picture by hand.¹⁶ The image will not look back on its own. But it can be arranged, by a careful writer, to be a picture in which something is looking. That is enough.

VI. Practical Coda — Prompting a Portrait That Looks Back

I want to close with three concrete practices I have been testing in workshop classes, and then with three Midjourney prompts that may help. None of these is a solution. They are practices for the long apprenticeship of putting a gaze back into a machine-made image.

First, refuse the default eye contact. The portrait conventions available to contemporary image models are saturated with sitters who look at or near the camera. The smooth result, requested without qualifiers, often offers the same. Begin your prompt with the angle of refusal. A woman whose face has just turned a quarter past the camera, her gaze leaving the lens by a few degrees.A man looking past the photographer to something the photographer cannot see.A child whose eyes are closed in concentration, not for sleep. When eye contact is refused, the picture can no longer pretend that the eye and the gaze sit in the same place. The viewer is forced to look for the gaze elsewhere, and the picture begins to organize a place where the gaze might be.

Second, install a structural exception. Holbein's skull interrupts the dominant frontal order of The Ambassadors; it is fully part of the painting's memento mori structure, but it becomes legible only from a place that disturbs the painting's official view. AI portrait prompts, almost by reflex, want to be coherent. Coherence can kill the gaze. Insert one thing that does not comfortably belong: a reflection in a window that should not be there, a hand reaching into the frame from outside the body of the sitter, a smear on a lens that the model is asked to render. The smear is not a flaw. It is the place from which the picture begins to look back.

Third, give the picture a witness. The photographic portrait carries an implicit second person — the photographer, the person who was there. The AI portrait, by default, has no one. You can install one by writing the witness into the prompt: seen from the angle of a person standing one step too close, photographed by a sister, not a stranger, taken in the moment after the conversation has stopped. The witness will not appear in the picture. But the picture will be arranged as if a witness were there, and the viewer, picking up the geometry, will fill the place. This is how the gaze is built — not from the front, but from the side, by the geometry of someone who was supposed to have been looking.

To make these practices concrete, three Midjourney prompts.

Prompt 1 — The Refused Eye

portrait of a woman in her late fifties, head turning a quarter past the camera as if a sound has just reached her from the left, soft north-window light from a single tall window, the eyes leaving the lens by a few degrees, lower face in slight shadow, photographed by a daughter standing one step too close, 1970s European domestic interior, hand-printed silver gelatin texture, slight grain --ar 4:5 --style raw

Intent: The gaze is constructed by the refused eye contact. The witness ("a daughter") installs the geometry of a returned look without rendering the witness herself.

Prompt 2 — The Holbein Stain

formal portrait of an elderly man seated in a study, three-quarter view, hands folded on his knee, looking past the photographer toward an unseen interlocutor, a long mirror at the back of the room reflects a shape that is almost but not quite a face turned away, the reflection slightly out of perspective, oil-on-canvas texture, Northern Renaissance studio light, vanitas mood, the surface of the desk holding a single overturned glass --ar 4:5 --style raw

Intent: The mirror introduces an anamorphic second perspective. The sitter occupies the official position; the reflection occupies a position the sitter cannot see. The picture has a place from which it looks back.

Prompt 3 — The Empty Mirror Stage

portrait of a young person seen from behind, head and shoulders, standing in front of a bathroom mirror in early morning light, the mirror is fogged so that the reflection is only a soft warmth where the face would be, the viewer is positioned slightly off-axis so that the reflected face cannot resolve, soft cyan light from a small window, breathing visible on the cold surface, the image holds the moment before the face is wiped clear --ar 4:5 --style raw

Intent: The mirror stage is structurally present and structurally refused. The sitter is identified with an image that the picture itself has withheld.

---

Lacan said, in the seminar where this essay began, that we are pictures.¹⁷ I take this to mean that we are not the masters of the visual field but elements arranged within it by a gaze we did not authorize. Fully synthetic AI portraiture, by removing the body that once stood in front of the lens, has stripped the field of its returned gaze and left us with a magnificent average. The temptation, in front of these images, is to assume that the gaze has gone away. It has not gone away. It has been waiting, on the side of the practitioner, to be put back into the picture by a careful arrangement of the field. The image cannot look back by itself. The writer can build a place from which it begins to.

The candle, as I said in an earlier essay, has not yet gone out.¹⁸ Neither has the gaze. It is waiting at the side of the frame, at the angle from which a skull resolves into a skull. The work of the practitioner is to walk to that angle and stay there for a while, and to write the picture as if someone were already looking back.

---

Tags

`gaze` · `lacan` · `ai-portrait` · `holbein` · `objet-petit-a` · `mirror-stage` · `eldagsen` · `pseudomnesia` · `promptography` · `composite-memory` · `prompt-design`

Related Essays on This Blog

- Composite Memory: How AI Rewrites What We Remember

- Stimmung: Why "Moody" Is Not a Prompt Strategy

- The Latent and the Manifest: Freud, Galton, and the Logic of AI Images

- The Texture of Time: Tarkovsky's Candle and the AI Image

Notes

1. Jacques Lacan, The Four Fundamental Concepts of Psychoanalysis, ed. Jacques-Alain Miller, trans. Alan Sheridan (New York: W. W. Norton, 1978), 106. The seminar was delivered in 1964; the French text was established by Jacques-Alain Miller and published as Le Séminaire, Livre XI: Les quatre concepts fondamentaux de la psychanalyse (Paris: Seuil, 1973).

2. The four sessions on the gaze in Seminar XI are gathered under the heading "Of the Gaze as Objet Petit a" and span chapters 6 through 9 of the published English translation. Lacan, Four Fundamental Concepts, 65–119.

3. On Lacan's scopic field, the shift from mirror to anamorphosis, and the relation among le regard, Barthes, and photographic affect, see Margaret Iversen, Beyond Pleasure: Freud, Lacan, Barthes (University Park, PA: Pennsylvania State University Press, 2007), esp. the introduction, "From Mirror to Anamorphosis," and the later chapters on photography and Camera Lucida.

4. The Petit-Jean anecdote is told in Lacan, Four Fundamental Concepts, 95.

5. Joan Copjec, Read My Desire: Lacan Against the Historicists (Cambridge, MA: MIT Press, 1994), esp. ch. 2, "The Orthopsychic Subject: Film Theory and the Reception of Lacan." The chapter was first published in October 49 (Summer 1989).

6. Bruce Fink, The Lacanian Subject: Between Language and Jouissance (Princeton, NJ: Princeton University Press, 1995), esp. ch. 7, "Object (a): Cause of Desire," 83–97.

7. Hal Foster, The Return of the Real: The Avant-Garde at the End of the Century (Cambridge, MA: MIT Press, 1996). Foster's most concentrated engagement with Lacan's gaze and the "return of the real" in late twentieth-century art appears in the chapter "The Return of the Real," which echoes the volume's title; closely related material was published as "Obscene, Abject, Traumatic," October 78 (Autumn 1996): 106–24.

8. Roland Barthes, Camera Lucida: Reflections on Photography, trans. Richard Howard (New York: Hill and Wang, 1981), 76–77. Originally published as La chambre claire: Note sur la photographie (Paris: Cahiers du Cinéma / Gallimard / Seuil, 1980). The concept of ça-a-été — "that-has-been" — is developed throughout the second half of the book.

9. Hans Holbein the Younger, Jean de Dinteville and Georges de Selve ("The Ambassadors"), 1533, oil on oak, 207 × 209.5 cm, National Gallery, London, NG1314. National Gallery catalogue entry: <https://www.nationalgallery.org.uk/paintings/hans-holbein-the-younger-the-ambassadors>. Lacan's reading of the anamorphic skull appears in Four Fundamental Concepts, 85–90.

10. On anamorphosis, see Jurgis Baltrušaitis, Anamorphic Art, trans. W. J. Strachan (New York: Harry N. Abrams, 1977). Originally published as Anamorphoses ou perspectives curieuses (Paris: Olivier Perrin, 1955). For the memento mori and National Gallery account of the skull in The Ambassadors, see the National Gallery catalogue entry for NG1314: <https://www.nationalgallery.org.uk/paintings/hans-holbein-the-younger-the-ambassadors>.

11. Lacan's most concentrated formulations of objet petit a in relation to the scopic field are in Four Fundamental Concepts, 67–78 and 103–7. For a concise secondary account, see Fink, The Lacanian Subject, esp. ch. 7.

12. Boris Eldagsen, PSEUDOMNESIA | The Electrician, 2022, AI-generated / AI-assisted image, from the series PSEUDOMNESIA — Fake Memories. The image was selected as winner of the Sony World Photography Awards 2023, Open Competition, Creative Category. Eldagsen refused the prize at the London award ceremony on 13 April 2023; the artist's own statement and chronology are published at <https://www.eldagsen.com/sony-world-photography-awards-2023/>.

13. Eldagsen credited Peruvian photographer Christian Vinces with suggesting the term promptography in the days immediately after the refusal; Eldagsen then adopted it as a cleaner term for AI-generated images using photographic language. Eldagsen's own framing, "I call my images 'images'… They are synthetically produced, using 'the photographic' as a visual language. They are not 'photographs,'" appears in the same statement: <https://www.eldagsen.com/sony-world-photography-awards-2023/>.

14. Jacques Lacan, "The Mirror Stage as Formative of the I Function as Revealed in Psychoanalytic Experience," in Écrits: The First Complete Edition in English, trans. Bruce Fink in collaboration with Héloïse Fink and Russell Grigg (New York: W. W. Norton, 2006), 75–81. The version collected in Écrits was delivered in Zurich in 1949; Lacan's first presentation of mirror-stage material dates back to 1936.

15. Hito Steyerl, "Mean Images," New Left Review 140/141 (March–June 2023): 82–97, <https://newleftreview.org/issues/ii140/articles/hito-steyerl-mean-images>. For a fuller treatment of how AI images inherit and recombine cultural memory — what I have called composite memory — see "Composite Memory: How AI Rewrites What We Remember" on this blog.

16. Caspar David Friedrich, Wanderer above the Sea of Fog / Der Wanderer über dem Nebelmeer, c. 1817/18, oil on canvas, 94.8 × 74.8 cm, Hamburger Kunsthalle, Hamburg. Hamburger Kunsthalle online collection: <https://online-sammlung.hamburger-kunsthalle.de/en/objekt/HK-5161>. The Rückenfigur device — the figure seen from behind — installs a place for the viewer's gaze that is neither simply inside nor outside the picture.

17. Lacan, Four Fundamental Concepts, 106. The full sentence is the epigraph of this essay.

18. The closing image is borrowed from "The Texture of Time: Tarkovsky's Candle and the AI Image" on this blog.

The Texture of Time

Tarkovsky's Candle and the AI Image

"Rhythm in cinema is conveyed by the life of the object visibly recorded in the frame. Just as from the quivering of a reed you can tell what sort of current, what pressure there is in a river, in the same way we know the movement of time from the flow of the life-process reproduced in the shot." — Andrey Tarkovsky, Sculpting in Time¹

I. A Man Carries a Candle

A man crosses an empty pool, cupping a candle in his hands. He walks slowly. Very slowly. The camera keeps patient measure with him. The flame wavers. It goes out. He walks back, relights it, begins again. And finally, at the end of a single unbroken take that lasts some nine minutes, he reaches the other side. He sets the candle on the pool's edge. And then he falls.

This is the climactic candle sequence near the end of Andrey Tarkovsky's Nostalghia (1983), filmed in the drained mineral pool of Bagno Vignoni in Tuscany's Val d'Orcia.² Oleg Yankovsky, playing the exiled Russian poet Andrei Gorchakov, carried the candle from one side to the other in a single take — no cut, no stitch, no cinematic sleight of hand. As Yankovsky later recalled, Tarkovsky had told him that if the act could be completed in one shot, straight, without tricks or editing, then perhaps this would be the true meaning of his life.³

Nothing happens in those nine minutes. There is almost no dialogue. There is no event, in any ordinary sense of the word. And yet anyone who has watched this scene remembers that for those nine minutes their own breath was tied to a small flame. Every time the candle wavered, the audience stopped breathing with it. This is not a matter of taste. It is a physiological event.

If you have seen this scene, you know what I am speaking of. If you have not, there is no way to translate it into words. You can describe it. But the gap between description and experience is precisely what Tarkovsky spent his life trying to protect. He gave that gap a name. He called it time.

II. Sculpted Time

In Sculpting in Time, Tarkovsky defined the essence of cinema as the act of sculpting time itself.⁴ A painter works with pigment. A sculptor works with stone. A musician works with sound. For the film director, Tarkovsky said, the material is time. The question was never what to put inside the frame but how time should flow within it. That was cinema, for him. That was everything.

This was a direct argument against a dominant twentieth-century theory of montage. Eisenstein had taught that meaning is born in the collision between images — in the cut, the splice, the dialectical shock of one shot against another. The grammar of cinema was editing. Tarkovsky refused this. For him, meaning was not born between shots but within them, in what he called "time-pressure," the density of time already flowing inside a single unbroken take.⁵ This is why his camera lingers. It waits for the candle to go out. It waits for the drop to fall. It waits for the wind to pass through the grass and keep going.

Many attempts have been made to explain this concept academically. Gilles Deleuze read Tarkovsky as a central figure in the post-war shift from the "movement-image" to the "time-image," the moment when European cinema stopped subordinating time to action and began to show time directly.⁶ Deleuze's reading rests on Henri Bergson's concept of durée — the claim that lived time is not a series of spatial units but a qualitative flow that cannot be cut into pieces without losing what it is.⁷ These are valid readings. But in this essay I want to stay one step before the theory, at the place Maurice Merleau-Ponty called description: not explanation or analysis, but the careful return to experience itself.⁸ The texture of time is not explained. It is felt.

III. How Matter Holds Time

There is always matter in a Tarkovsky frame. Water, fire, earth, wind, damp walls, old wood. Robert Bird, in his study of the filmmaker, arranges Tarkovsky's entire body of work under the four classical elements — water, fire, earth, air — and the arrangement is not a conceit but a testimony.⁹ Tarkovsky's camera was turned toward matter. It watched how matter was enduring time.

In a scene from Mirror, wind sweeps across a field and bends the grass. Nothing happens. And yet, after the wind has passed, we feel that this field has been standing here for decades. In Stalker, water moves slowly across the floor of a room. The water says nothing. And yet, in its slow movement, we know that this room has been empty for a long time. Tarkovsky is not showing us a field, or a room, or a candle. He is showing us the time that has already passed through them, and the time that is still passing.

The texture of time lives in matter. A candle only has time when it has become "a candle that has endured nine minutes." A candle with no duration behind it has no time. This is why Tarkovsky insisted on the long take. Time has to accumulate before it becomes visible. Time that has not accumulated is not time. It is only a moment.

A question arises here. Can the texture of time — matter holding duration — be carried in a single frame, in one still image? This question has held me for a while. Since the invention of photography, the still image has always wrestled with it. Henri Cartier-Bresson's "decisive moment" was one answer: the instant in which all the forces of a scene converge into a recognizable pattern.¹⁰ Roland Barthes, in Camera Lucida, offered another: the punctum, the small detail that pierces the surface of the image and pulls the viewer out of ordinary time into the presence of something that has been.¹¹ These are different answers to the same question. Each great photographer, Tarkovsky might have said, has been sculpting time with a shutter.

IV. Before the AI Image

Now I stand before an image generated by an AI. There is a man. There is an empty pool. There is a candle in his hand. The light is beautiful. The composition is precise. The color is cinematic. Everything is there. And yet my breath does not stop.

What is missing? It is difficult to say. Something you cannot quite call present or absent. I have been thinking about this for a long time, and for now this is the only way I can put it: the candle in that image is not a candle that could have gone out. The candle has always been exactly as it is. There is no "a moment ago" inside that image, and no "a moment from now." There is only "now." But what makes Tarkovsky's candle stop our breath is precisely that "a moment ago" and "a moment from now" — the knowledge that the candle has already almost failed, and that therefore it could still fail. That is what gives the candle its time.

I feel the temptation to explain this mechanistically. Why does the AI image fail to hold time? What is the structure of the model that generates it? What rises from the latent space?¹² These are questions for another kind of writing. I will not open that door here. The task of an essay is not to explain a mechanism but to describe a phenomenon accurately. And here is what I can describe right now: AI images are too good at rendering the completed moment. So good, in fact, that there is no room left inside the image for time to settle in. AI can already reconstruct atmospheric residue and emotional weight — what it still rarely holds is duration, the sense that something has been enduring before we arrived.¹³

V. How to Invite Time Into a Prompt

Should we then give up? No. This is not a limit of the medium. It is a sign that we have not yet learned the language of the medium. Cinema existed before Tarkovsky discovered sculpted time. It simply did not yet know how to hold time. The AI image is likely in the same stage now, and the people who will teach it are the writers — because this is a medium that, for the first time in the history of images, is shaped by sentences.¹⁴

Let me share three practices I have been testing, in my workshop classes and in my own work. They are not solutions. They are ways of inviting time into the prompt.

First, describe a state in progress, not a state completed. Not "a candle," but "a candle that has just wavered and settled again." Not "a wall," but "a wall through which rainwater has been seeping for decades." AI models are superb at generating finished states; what they need is for the state to be specified as one point in a process. When you do this, a trace of time begins to show on the surface of the image.

Second, name the history of the light. Not "afternoon light," but "light that has been resting on this windowsill since morning." Not "a shadow," but "a shadow slowly lengthening as the sun tilts down." Light is the clearest carrier of time, and when you write how it has endured, the model tries to reflect that endurance on the surface.

Third, give the matter an age. Not "a wooden table," but "the wooden table my father used, its surface slightly worn where his elbow rested." Not "a glass," but "a glass whose rim is polished smooth because the same hand has held it in the same place every day." When matter has age, matter holds time.

These are not formulas. The moment they become formulas, they turn into another cliche. They are only a starting point for a writer who is trying to carry time into an image. Each person's texture of time must be discovered in their own language.

VI. The Candle That Endures

I return to Tarkovsky's candle. Andrei Gorchakov carried it to the other side. That is all that happened. And then, as soon as he set it down, he fell. For those nine minutes, his life had been tied to the life of the flame. And for those same nine minutes, the breath of anyone watching had been tied to it as well. On the surface of that small fire, a man's last gesture and a stranger's perception met.

Can my image become a candle that endures nine minutes? I do not have the answer yet. But I believe that holding the question — each time I release the shutter, each time I write a prompt — is itself the way toward the answer. Tarkovsky's gift to us was not the command to sculpt time. It was the invitation to look at where time is already resting. That invitation is still open. It is open in cinema. It is open in photography. And it is open in the AI image — a medium still looking for its own language.

The candle has not yet gone out.

Related Essays on This Blog

• Composite Memory: How AI Rewrites What We Remember https://www.luxlatens.com/blog/composite-memory-how-ai-rewrites-what-we-remember

• Stimmung: Why "Moody" Is Not a Prompt Strategy https://www.luxlatens.com/blog/stimmung-why-moody-is-not-a-prompt-strategy

• A forthcoming essay: The Latent and the Manifest: Freud, Galton, and the Logic of AI Images

Notes

1. Andrey Tarkovsky, Sculpting in Time: Reflections on the Cinema, trans. Kitty Hunter-Blair (Austin: University of Texas Press, 1987), 119.

2. Nostalghia, directed by Andrey Tarkovsky (1983), starring Oleg Yankovsky. The candle sequence was filmed in the drained mineral pool of Bagno Vignoni, in the Val d'Orcia region of Tuscany, in a single take of nine minutes and four seconds.

3. Oleg Yankovsky's recollection of Tarkovsky's challenge is quoted in Robert Bird, Andrei Tarkovsky: Elements of Cinema (London: Reaktion Books, 2008), 192.

4. Tarkovsky, Sculpting in Time, esp. the chapters "Imprinted Time" and "The Film Image."

5. Ibid., 117–20. The concept of "time-pressure" (or "time-thrust") is central to Tarkovsky's theory of rhythm and is concentrated in these pages.

6. Gilles Deleuze, Cinema 2: The Time-Image, trans. Hugh Tomlinson and Robert Galeta (Minneapolis: University of Minnesota Press, 1989). Originally published as Cinéma 2: L'image-temps (Paris: Éditions de Minuit, 1985).

7. Henri Bergson, Time and Free Will: An Essay on the Immediate Data of Consciousness, trans. F. L. Pogson (London: S. Sonnenschein & Co., Ltd.; New York: The Macmillan Co., 1910). Originally published as Essai sur les données immédiates de la conscience (Paris: Alcan, 1889). A widely available modern reprint is the Dover edition (Mineola, NY: Dover, 2001).

8. Maurice Merleau-Ponty, Phenomenology of Perception, trans. Donald A. Landes (London: Routledge, 2012), Preface. Originally published as Phénoménologie de la perception (Paris: Gallimard, 1945).

9. Robert Bird, Andrei Tarkovsky: Elements of Cinema (London: Reaktion Books, 2008). Bird organizes his study around the four classical elements of water, fire, earth, and air.

10. Henri Cartier-Bresson, The Decisive Moment, in collaboration with Éditions Verve of Paris (New York: Simon and Schuster, 1952); originally Images à la sauvette (Paris: Éditions Verve, 1952).

11. Roland Barthes, Camera Lucida: Reflections on Photography, trans. Richard Howard (New York: Hill and Wang, 1981). Originally published as La chambre claire: Note sur la photographie (Paris: Cahiers du Cinéma / Gallimard / Seuil, 1980). The concept of the punctum is developed in the second half of the book.

12. For a detailed account of the latent/manifest model across photography, psychoanalysis, and AI, see a forthcoming essay, "The Latent and the Manifest: Freud, Galton, and the Logic of AI Images," on this blog.

13. For a fuller account of how AI images inherit and recombine cultural memory — including what I call composite memory — see "Composite Memory: How AI Rewrites What We Remember" on this blog. https://www.luxlatens.com/blog/composite-memory-how-ai-rewrites-what-we-remember

14. On the practical distinction between atmospheric adjectives like "moody" and the material conditions that actually produce Stimmung in a prompt, see "Stimmung: Why 'Moody' Is Not a Prompt Strategy" on this blog. https://www.luxlatens.com/blog/stimmung-why-moody-is-not-a-prompt-strategy

Stimmung: Why "Moody" Is Not a Prompt Strategy

On Atmosphere, Attunement, and the Material Conditions of Mood in Photography and AI

Type "moody" into Midjourney and you will get an image: desaturated, low-key, probably foggy, probably featuring a small figure in a vast landscape. It will look atmospheric. It will look like every other "moody" image ever generated. This is not Stimmung. This is the statistical average of Stimmung — Hito Steyerl's "mean image" applied to mood itself.¹

The German word Stimmung has no adequate English translation. It encompasses mood, atmosphere, attunement, and tuning — all at once. Its etymology is revealing: Stimmen means to tune a musical instrument. From the physical act of bringing strings into correct pitch, the word expanded to describe the tuning of the mind, the atmosphere of a place, the mood of an era. Its philosophical history — from Kant through Heidegger to contemporary aesthetics — offers a framework that transforms how we think about the "atmosphere" we chase in our images.

This essay traces that framework and argues that for photographers and AI visual artists, Stimmung is not an adjective to type into a prompt but a material condition to design.

1. What Stimmung Is (and What "Moody" Is Not)

The uniqueness of Stimmung lies in its double reach: both subjects and objects can be "in" it. "I am melancholic" describes a subject's mood. "This room is melancholic" describes an object's atmosphere. The word covers both without distinguishing them — because in the phenomenological tradition that gave Stimmung its deepest philosophical articulation, the distinction itself is the problem.²

Immanuel Kant was the first to give Stimmung a formal role in aesthetic theory. In the Critique of Judgment (1790), he argued that the experience of beauty arises from a "free play" (freies Spiel) between imagination and understanding — a harmonious attunement of our cognitive faculties that precedes any conceptual judgment. This state of attunement is what Kant calls aesthetic Stimmung.³ The implication for image-making: what we call a "good image" often elicits not a specific idea or emotion but a felt sense of being attuned — something beyond content that we recognize but struggle to name.

German Romantic painting gave this concept its first visual form. Caspar David Friedrich (1774–1840) invented the visual vocabulary of Stimmung: fog-shrouded mountains, Gothic ruins, solitary figures seen from behind (Rückenfigur) gazing into vast landscapes. The Rückenfigur functions as the viewer's proxy — a device that draws our gaze not at the landscape but into it. Friedrich's paintings do not depict scenery; they engineer immersion.⁴ J.M.W. Turner (1775–1851) took this further: in his late works, light and atmosphere dissolve the contours of objects entirely. Stimmung becomes not an attribute of things but an energy field — a move that anticipates Abstract Expressionism and, two centuries later, the "atmosphere-first" logic of AI image generation.

2. Heidegger: Stimmung as the Ground of Being

Martin Heidegger radically transformed Stimmung from a psychological category into an ontological one. In Being and Time (1927) and The Fundamental Concepts of Metaphysics (1929/30), he argued that Stimmung is not an emotion that arises inside us and then colors our perception of the world. It is the prior condition through which we encounter the world at all — what he calls Befindlichkeit, our fundamental "findingness" or "how we find ourselves" in any given situation.⁵

For Heidegger, mood is not an interior feeling later projected outward; it pervades our being-in-the-world from the start.⁵ As he writes in The Fundamental Concepts of Metaphysics: "A person who 'brings spirit to a group' does not produce a psychic experience internally and then transfer it outward like an infectious germ. Rather, the mood is, so to say, an atmosphere in which we are steeped and by which we are thoroughly attuned."⁶

For photography, this reframing is consequential. The "atmosphere" of a good photograph is not something the photographer "puts in." It is something already present in the relationship between scene, light, time, and observer — something the photographer discovers and captures. For AI image generation, the parallel is equally significant: you cannot command Stimmung through a prompt. You can only activate patterns of Stimmung that already exist latently within the statistical distributions of the training data.

3. Gumbrecht and Böhme: Presence and the Designability of Atmosphere

Hans Ulrich Gumbrecht brought Stimmung back into literary and aesthetic theory with Atmosphere, Mood, Stimmung: On a Hidden Potential of Literature (2012). His central argument: the function of art is not to represent (re-present) but to make present. Reading literature — or experiencing any artwork — is not primarily an act of interpretation but an encounter with a specific Stimmung that physically envelops us. "Reading for Stimmung," Gumbrecht writes, "always means paying attention to the textual dimension of the forms that envelop us and our bodies as a physical reality — something that can catalyze inner feelings without matters of representation necessarily being involved."⁷

Gumbrecht describes Stimmung as existing on "a continuum akin to a musical scale" — nuances that challenge our powers of discernment and description. Perhaps the best we can do is point in their direction. This is both the essential limitation and the essential dignity of the prompt: it is a gesture toward an atmospheric quality that language can only approximate.

Gernot Böhme, working from Hermann Schmitz's neo-phenomenology, takes the argument in a more practical direction. In Atmospheric Architectures (2017), Böhme defines atmosphere as neither a subjective emotion nor an objective property but a relational phenomenon between subject and environment. Crucially, he argues that atmosphere is designable (gestaltbar). Architecture, lighting, music, and weather produce atmospheres not by accident but through deliberate material arrangement.⁸

This is the philosophical license for what we do with prompts. If Stimmung is a designable relational phenomenon produced by specific material conditions, then the prompt is — or should be — an atmospheric design document. Not "moody." Not "cinematic." But a specification of material conditions (light, atmosphere, space, temporal traces, color) from which a particular Stimmung will emerge.

4. The Materiality of Stimmung: Five Layers

Stimmung is not an abstraction. It arises from concrete material conditions. In photography and AI image generation, five material layers constitute the building blocks of atmosphere:

Light — time of day (golden hour, blue hour, midnight), direction (backlit, sidelit, toplit), quality (harsh, soft, diffused, dappled), and color temperature (warm, cool, neutral). Light is the primary carrier of Stimmung; change the light and you change everything.

Atmosphere — fog, mist, haze, rain, drizzle, dust, smoke, humidity, condensation. The medium through which light travels determines how it behaves — whether it dissolves contours (Turner), filters through glass (Sudek), or creates shafts and shadows (Fan Ho).

Space — depth (deep, shallow, compressed), scale (vast, intimate, claustrophobic), the ratio between figure and environment, and boundaries (windows, frames, obstructions). Gaston Bachelard's Poetics of Space (1958) showed that spatial archetypes carry inherent Stimmung: the attic is a space of solitary thought; the cellar, of unconscious dread.⁹

Temporal traces — weathering, rust, patina, fading, seasonal markers (fallen leaves, first snow, dried grass), signs of abandonment or absence. Maurice Merleau-Ponty's phenomenology of embodied perception reminds us that Stimmung is fundamentally synaesthetic — we "feel" the dampness of a ruin, the cold of a foggy morning, the silence of an empty room, even in a flat image on a screen.¹⁰

Color — saturation (muted, saturated, desaturated), palette (monochrome, limited, complementary), tonal range (high-key, low-key, mid-tone), and specific chromatic shifts (sepia, cyan, warm cast). Wilhelm Hammershøi (1864–1916), the Danish painter of silent interiors, worked with no more than eight to ten colors — proof that a severely restricted palette can intensify rather than diminish Stimmung.¹¹

The critical shift for prompt design: replace atmospheric adjectives with material specifications. Not "moody landscape" but "early March, just after rain, residual golden light on wet asphalt, thin layer of ground-level mist, neon reflections in puddles, muted three-color palette."

5. Twelve Masters of Stimmung

The following photographers and artists represent distinct positions on the spectrum of Stimmung. Each demonstrates a different principle of how atmosphere is materially constructed.

Andrey Tarkovsky (1932–1986) — filmmaker, but an indispensable reference for photographic Stimmung. In Nostalghia (1983), Stalker (1979), and Mirror (1975), Stimmung arises from slow time: water flowing, grass swaying, light moving across walls. For Tarkovsky, the director's task was to make time itself perceptible within the shot — not through faster action, but through duration, rhythm, and the pressure of life unfolding in matter.¹² For still images, the challenge is to imply this temporal texture — the sense that time has passed slowly through a scene — without motion.

Wilhelm Hammershøi (1864–1916) — painted silence as Stimmung. Gray Copenhagen interiors, a woman seen from behind, empty rooms, dust-filtered light. Sound is entirely absent, and this absence itself becomes a palpable atmosphere. Hammershøi proves that Stimmung can emerge from subtraction rather than addition.¹³

Josef Sudek (1896–1976) — Prague's "poet of photography." His studio window series — foggy glass, condensation, blurred gardens beyond — demonstrates that Stimmung is experienced through mediated vision. The window is both a physical boundary and a Stimmung filter.¹⁴

Fan Ho (1931–2016) — light as protagonist. In 1950s–60s Hong Kong, Ho turned alleys and staircases into theaters of dramatic chiaroscuro. Figures become silhouettes within geometric light, transforming urban everyday life into staged scenes.¹⁵

Michael Kenna (1953–) — long exposures at dawn or dusk; fog, snow, still water. Time is compressed into a single image. A lone tree or pier in vast empty space. Extreme minimalism amplifies Stimmung: less is more.¹⁶

Hiroshi Sugimoto (1948–) — the Seascapes reduce the world to a horizon dividing sky and sea. The Theaters expose an entire film's running time onto a single frame, producing a blank white screen — two hours of narrative compressed into pure light. Sugimoto converts time itself into Stimmung.¹⁷

Masao Yamamoto (1957–) — tiny prints, deliberately stained and creased, carried in pockets. Yamamoto's images are meant to be touched, not just seen. This haptic Stimmung — the sense of material contact — is precisely what digital and AI images most fundamentally lack.¹⁸

Rinko Kawauchi (1972–) — a rare practitioner of bright Stimmung. Light glinting on insect wings, water droplets, sparks — the sacred discovered in the micro-moments of ordinary life. High-key tones, shallow depth of field, overexposure at the edge of dissolution. Kawauchi proves that Stimmung does not require darkness.¹⁹

Edward Hopper (1882–1967) — the paradox of loneliness in full light. Nighthawks (1942), Morning Sun (1952): warm illumination that isolates rather than connects. Visibility without intimacy. This contradictory Stimmung — light plus solitude — is the signature mood of modern urban existence.²⁰

Saul Leiter (1923–2013) — blur, obstruction, color fields. Leiter refused sharp focus and the decisive moment, instead capturing New York through rain, reflections, and out-of-focus foregrounds. The city is not recorded but felt.²¹

Luigi Ghirri (1943–1992) — everyday metaphysics. Clotheslines, road signs, beaches, walls — unremarkable subjects in muted color, balanced composition, and generous empty space. Ghirri demonstrated that strong Stimmung requires neither dramatic lighting nor atmospheric effects; it can emerge from the quiet contemplation of ordinary things.²²

Alec Soth (1969–) — narrative Stimmung across a series. Sleeping by the Mississippi (2004) accumulates its melancholy not through single images but through the arc of an entire book: the vast spaces, declining communities, and transient dwellings of the American interior. Soth shows that Stimmung is not only an attribute of individual images but can be built across a body of work.²³

6. Stimmung and AI: The Cliche Problem

Midjourney is remarkably competent at generating Stimmung — and remarkably limited. Prompt words like "moody," "atmospheric," "ethereal," and "cinematic" activate strong visual patterns in the latent space. But these patterns are, as David Bate's analysis of composite memory demonstrates, statistical averages of cultural value.²⁴ "Moody" activates the intersection of millions of images tagged or captioned with that word, converging on a predictable mean: dark, desaturated, foggy, small figure, blue-gray color temperature.

This is the Stimmung equivalent of what Bate, following Francis Galton's composite photography, identifies as the erasure of individuality through statistical averaging.²⁵ Steyerl calls these outputs "mean images" — representations of the statistical average rate in the particular.²⁶ Palmer and Sluis identify the broader phenomenon as "the automation of style."²⁷

Four strategies can break this convergence:

Specificity — Replace adjectives with material conditions. Böhme's aesthetics of atmospheres teaches that mood is a relational phenomenon arising from material arrangements, not an emotion to be named.²⁸ Not "moody" but "March, Seoul, 6:40 AM, wet asphalt after rain, neon reflections in puddles, thin ground mist."

Contradiction — The power of Stimmung often lies in paradox. Hopper's warm light plus loneliness. Kawauchi's brightness plus the sacred. Tarkovsky's ruins plus awe. Replace single mood adjectives with two contradictory atmospheric conditions.

Conscious reference — Using --sref or "in the style of" is not inherently problematic, but it requires knowing why a given artist's Stimmung works — what material and historical conditions produced it. Bate warns about the "de-historicization of style" when historical aesthetics become detachable algorithmic parameters.²⁹

Subtraction — Hammershøi and Kenna demonstrate that Stimmung intensifies through removal: fewer modifiers, more empty space, no narrative, restricted palette. Bernard Stiegler's warning about digital automation "short-circuiting the deliberative functions of the mind"³⁰ is relevant here: the impulse to add more prompt words is precisely the reflex that produces cliches.

7. Conclusion: Finding Your Own Stimmung

The theoretical arc from Kant to Böhme establishes that Stimmung is not decoration or afterthought but the foundational condition of aesthetic experience. The twelve masters demonstrate that atmospheric mastery operates through precise material choices, not vague adjectives. And the analysis of AI's statistical convergence reveals why "moody" will never produce what Kenna or Sudek or Tarkovsky achieved.

The task, finally, is not to reproduce another artist's Stimmung but to discover your own. What light do you respond to? In what kind of space does time seem to stop for you? What atmospheric conditions take your breath away? These questions are not sentimental. They are the core of artistic identity — and they cannot be answered by a statistical average.

Stimmung is not an adjective. It is a material condition. Design it.

Related Essays on This Blog

• Composite Memory: How AI Rewrites What We Remember https://www.luxlatens.com/blog/composite-memory-how-ai-rewrites-what-we-remember

• The Texture of Time: Tarkovsky's Candle and the AI Image

Notes

1. Hito Steyerl, "Mean Images," New Left Review 140/141 (2023): 82–97.

2. Otto Friedrich Bollnow, Das Wesen der Stimmungen (Frankfurt: Klostermann, 1953).

3. Immanuel Kant, Critique of the Power of Judgment (Cambridge: Cambridge University Press, 2000), originally published 1790.

4. Joseph Leo Koerner, Caspar David Friedrich and the Subject of Landscape (London: Reaktion Books, 1990).

5. Martin Heidegger, Being and Time, trans. Joan Stambaugh (Albany: SUNY Press, 1996), originally published 1927; and The Fundamental Concepts of Metaphysics: World, Finitude, Solitude, trans. William McNeill and Nicholas Walker (Bloomington: Indiana University Press, 1995), originally published 1929/30.

6. Heidegger, The Fundamental Concepts of Metaphysics, 67.

7. Hans Ulrich Gumbrecht, Atmosphere, Mood, Stimmung: On a Hidden Potential of Literature (Stanford, CA: Stanford University Press, 2012).

8. Gernot Böhme, Atmospheric Architectures: The Aesthetics of Felt Spaces (London: Bloomsbury Academic, 2017); and Tonino Griffero, Atmospheres: Aesthetics of Emotional Spaces (New York: Routledge, 2014).

9. Gaston Bachelard, The Poetics of Space (Boston: Beacon Press, 1994), originally published 1958.

10. Maurice Merleau-Ponty, Phenomenology of Perception, trans. Donald A. Landes (London: Routledge, 2012). Originally published as Phénoménologie de la perception (Paris: Gallimard, 1945).

11. Poul Vad, Vilhelm Hammershøi and Danish Art at the Turn of the Century (New Haven: Yale University Press, 1992).

12. Andrey Tarkovsky, Sculpting in Time: Reflections on the Cinema, trans. Kitty Hunter-Blair (Austin: University of Texas Press, 1987), 117–20.

13. Vad, Vilhelm Hammershøi and Danish Art at the Turn of the Century.

14. Anna Farova, "Josef Sudek: Poet of Prague," Aperture 117 (Winter 1990).

15. Fan Ho, Hong Kong Yesterday (Palo Alto: Modernbook Editions, 2006).

16. Michael Kenna, Retrospective Two (Tucson: Nazraeli Press, 2004).

17. Hiroshi Sugimoto, Hiroshi Sugimoto (Ostfildern: Hatje Cantz, 2005).

18. Masao Yamamoto, Small Things in Silence (Mexico City: RM, 2015).

19. Rinko Kawauchi, Utatane (Tokyo: Little More, 2001).

20. Gail Levin, Edward Hopper: An Intimate Biography (New York: Knopf, 1995).

21. Saul Leiter, Early Color (Göttingen: Steidl, 2006).

22. Luigi Ghirri, The Complete Essays 1973–1991 (London: MACK, 2016).

23. Alec Soth, Sleeping by the Mississippi (Göttingen: Steidl, 2004).

24. David Bate, "AI Photography and Composite Memory," photographies 19, no. 1 (2026): 125–144.

25. Francis Galton, "Composite Portraits," Nature 18 (1878): 97–100.

26. Steyerl, "Mean Images."

27. Daniel Palmer and Katrina Sluis, "The Automation of Style: Seeing Photographically in Generative AI," Media Theory 8, no. 1 (2024): 159–184.

28. Böhme, Atmospheric Architectures.

29. Bate, "AI Photography and Composite Memory"; Kathrin Yacavone, "Virtual Photographs, Possible Memories: AI Images, the Archive, and the Works of August Sander and Elena Efeoglou," photographies 19, no. 1 (2026): 59–79; and Palmer and Sluis, "The Automation of Style."

30. Bernard Stiegler, Automatic Society, Volume 1: The Future of Work, trans. Daniel Ross (Cambridge, UK: Polity, 2016); cited in Bate, "AI Photography and Composite Memory," 134.

Composite Memory: How AI Rewrites What We Remember

A Photographer's Guide to the Theory Behind Generative Images

When Boris Eldagsen submitted an AI-generated image to the Sony World Photography Awards in 2023 and later refused the prize,¹ the media reduced the event to a simple binary: real photograph versus fake AI. But David Bate, in his landmark essay "AI Photography and Composite Memory,"² argues that this framing misses the point entirely. AI images are neither fake nor real. They are, in his formulation, "statistical samples, aggregates of cultural value made manifest in an image" — a new category that demands an entirely new theoretical vocabulary.

This essay traces that vocabulary through the recent special issue of photographies (Volume 19, Issue 1, March 2026), titled "Photography & Memory in the Age of AI," and connects it to the broader theoretical landscape that informs how we — as photographers and AI visual artists — should think about the images we make with generative tools.

1. The End of "That-Has-Been"

For over a century, photography's claim to truth rested on what Roland Barthes called the "that-has-been" (ça-a-été) — the physical trace of light on a photosensitive surface, guaranteeing that something was once there before the lens.³ AI-generated images shatter this guarantee. Nothing was ever in front of any camera. Instead, these images are statistical renderings extracted from latent spaces — mathematical compressions of millions of training images into high-dimensional vector fields.⁴

Fred Ritchin names this new image category "desirents" — visualizations not of "the way things are" but of "how one wants things to be," mapping what he calls "territories of unknown origin" within high-dimensional vector spaces.⁵ Kathrin Yacavone prefers the terms "virtual photographs" and "possible memories" — artifacts that visually mimic the aesthetic conventions of the camera while lacking any physical referent in the material world.⁶

These are not merely taxonomic distinctions. They reshape fundamental questions about evidence, memory, and cultural transmission — questions that become urgently practical the moment you open Midjourney and type a prompt.

2. The Psyche as Photographic Archive

Bate's theoretical point of departure is an insight from Jacques Derrida. In a 2000 lecture on photography, Derrida argued that human perception is already technical: "Within perception there are already selection, exposure time, filtering, and development." The psychic apparatus, Derrida claimed, functions "like" — or perhaps "as" — a photographic archive.⁷

The slippage between "like" and "as" is deliberate. Derrida suggests that the human mind and the photographic archive are not merely analogous; they share the same structural operations: selection, filtering, inscription. The question of inscription within the psychic apparatus had long been central to Derrida's engagement with Freudian psychoanalysis.⁸ Bate builds directly on this lineage. Just as every archive is selective — police archives select for criminal identification, family albums for happy moments, medical archives for health markers — human memory selects according to the topography of consciousness, preconsciousness, and the unconscious.⁹

The arrival of generative AI adds a third term to Derrida's analogy. If the psyche operates like/as a photographic archive, then AI's latent space constitutes a new kind of archive — one that is neither subjective (like memory) nor physically indexed (like photographs), but statistically aggregated from culture at large. Bate frames the central question: "If human perception cannot be separated from technical images, what are the consequences of generative AI images for human memory? Or conversely, what is the effect of human memory on this generative artificial image production model?"¹⁰

3. Latent / Manifest: A Model Across Three Domains

The most powerful conceptual tool in Bate's essay is the latent/manifest couplet, which he traces across three domains:

In analog photography, exposed film carries a latent image invisible to the eye until chemical development renders it manifest. In Freud's dream theory, latent thoughts (day residues, unconscious wishes) are transformed through the dreamwork — condensation, displacement, visual representation — into the manifest content of the dream.¹¹ In AI image generation, the latent space (statistical compressions of training datasets) is activated by a human prompt and processed through algorithmic computation (denoising, diffusion) into a manifest image.¹²

The parallel is not merely illustrative. Bate argues that all three domains share the same structural logic: visible sources are transformed through a process of selection and synthesis into visible images. The specifics of the process differ — chemistry, unconscious desire, algorithmic computation — but the latent-to-manifest arc is consistent. This connection matters because it situates AI image generation not as a rupture but as a continuation of existing models of memory and image-making.¹³

4. Galton's Ghost: Composite Photography and Its Legacy

This structural parallel reaches back further than either Freud or digital computing. In the 1870s, the statistician Francis Galton developed composite photography: a technique of superimposing multiple portrait photographs onto a single photographic plate through repeated partial exposures, producing a single "average" face that revealed "typical" features of a group.¹⁴

Galton himself noted that the composite portraits looked "better" than any of the individual constituents — precisely because the averaging process erased individual irregularities.¹⁵ When Midjourney generates a portrait, it too produces a statistically averaged face, one that tends to be smoother, more symmetrical, and more conventionally attractive than any real human face. The cliché is not a bug; it is a feature of statistical systems.

Galton's composites were later entangled with his eugenics project, and Allan Sekula's foundational critique "The Body and the Archive"¹⁶ established the terms for understanding how photographic archives can function as instruments of power and social control. Bate summons Galton not to rehabilitate eugenics but to expose the risks inherent in any system of statistical image generation: the erasure of individuality, convergence toward the mean, and the amplification of dataset biases.¹⁷ Catherine Malabou extends this genealogy from Galton through genetics, cybernetics, and epigenetics to contemporary AI, arguing that "the brain and the computer are in a reciprocal and mirroring relationship" and that naturalistic resistance to the technological capture of intelligence is meaningless.¹⁸

Freud himself cited Galton directly. In The Interpretation of Dreams, describing a dream figure that combined his uncle's face with a friend's yellow beard, Freud wrote: "It was like one of Galton's composite photographs."¹⁹ He called this process "condensation" — the compression of multiple latent elements into a single manifest image. Bate's central argument is that AI image generation performs a structurally identical operation, minus the human unconscious. The critical difference: in dreams, unconscious desire drives the condensation; in AI, "the human prompt gives semiotic activation and fire to the generative apparatus."²⁰

5. The Archive Transformed

If AI images are composite memories drawn from cultural archives, then the status of those archives matters enormously. Aleida Assmann defined the archive as a space of "passive remembering" — a repository where traces of the past remain in a "state of latency" until recalled.²¹ But generative AI transforms the archive from a passive storage facility into what Roland Meyer calls an "operative image archive": historical data is no longer preserved but "scraped" and "mined" to fuel neural network training. The archive's latency becomes commodified as generative potential.²²

Hito Steyerl names the outputs of this process "mean images" — visual representations of the statistical mean or average of training data.²³ These are more reductive than even Galton's composites, because they have no contact with physical reality whatsoever. Daniel Palmer and Katrina Sluis analyze this as "the automation of style," where a photographer's historically situated aesthetic is reduced to an algorithmic parameter.²⁴

Bernard Stiegler warned about the broader social consequences of such automation: "Digital automation short-circuits the deliberative functions of the mind," producing what he called "systemic stupidity" — a functionally drive-based mode of cultural production that replaces reflection with reflex.²⁵

6. Five Case Studies in Composite Memory

6.1 Boris Eldagsen: Reverse-Engineering Postmemory

Bate reads Eldagsen's The Electrician not through the media's "real vs. fake" narrative but as a case study in composite memory. Eldagsen's father, born in 1924, enlisted in the German army at seventeen and, like most of his generation, never spoke about the war. After his death, Eldagsen discovered photographs from the 1940s and began collecting similar images from flea markets and eBay. Using DALL-E 2, he synthesized new images that visualize the silent decade of his father's youth.²⁶

Bate connects this to Marianne Hirsch's concept of "postmemory" — the transmission of traumatic experience across generations through stories, images, and behaviors.²⁷ Generative AI offers the possibility of reverse-engineering this process: using historical images to generate new images of a past that was never visually documented. The resulting images are not "remembered" but "re-membered" — a bricolage of fragments that never originally belonged together, assembled into a composite that functions as artificial memory.²⁸

6.2 Ahn Jun: Materializing Verbal Memory

Korean artist Ahn Jun's 2023 project used Midjourney to generate 307 images published as a photobook in Japan. During his studies in early-2000s Los Angeles — before smartphones, before ubiquitous photography — he had no images of that period. He transformed stories and anecdotes heard from people he met into prompts, materializing "the life in California he had missed" as dreamlike images. The publisher describes this as "the materialization of imagined scenes through an AI image generator — itself a form of dreamlike imagination."²⁹

Bate reads both projects as "experiments in new memory-work using computers," analogous to the photo-therapy that predated the digital age — projects that invite audiences to "imagine other pasts, presents, and futures."³⁰

6.3 Elena Efeoglou: The De-historicization of Style

Elena Efeoglou's exhibition Blurring Reality and Fiction — August Sander meets AI (2025) reinterprets August Sander's People of the Twentieth Century through AI. Sander's original project was a visual taxonomy of Weimar-era German citizens classified by occupation and social role — a project whose printing plates were destroyed by the Nazis in 1936.³¹

Yacavone identifies three dynamics at work. First, individualization: Efeoglou assigns names and fictional biographies to Sander's anonymous subjects. Second, a counter-pull toward typification: AI's statistical averaging reasserts generic "types" despite the artist's efforts at individuation. Third, de-historicization of style: Sander's New Objectivity aesthetic — a product of specific social conditions in interwar Germany — becomes a detachable "metastyle" when translated into a Midjourney prompt parameter.³²

6.4 Exhibit A-i: From Evidence to Testimony

The Exhibit A-i: The Refugee Account project (2023) addresses human rights abuses in Australia's offshore detention centers on Nauru and Manus Island — sites where the government has systematically blocked physical access, making photographic documentation impossible. Based on testimonies from 32 refugees, the project used Midjourney to generate 130 images, which were then uploaded to Shutterstock alongside conventional photojournalistic assets.³³

Svea Braeunert's analysis centers on the shift from evidence to testimony. These images lack physical causality and cannot function as legal evidence. But they can function as testimony — a subjective, conceptual reconstruction of lived experience made visible. Notably, the AI-specific artifacts (distorted fingers, painterly textures, expressionistic distortion) function not as technical failures but as ethical devices that visualize unspeakable suffering.³⁴

Yet this raises Ariella Azoulay's concern about the "civil contract" of photography — the ethical interaction between subject, photographer, and viewer that traditional portraiture presupposes.³⁵ AI-generated images of vulnerable populations operate in the absence of this contract, as the Amnesty International controversy over AI-generated images of the Colombian protests made clear.³⁶

6.5 Eyes That Never Looked Back

Sara Oscar and colleagues give this theme its sharpest formulation. The eyes of AI-generated portraits are composite averages — statistical means synthesized from thousands of faces in the training data. In traditional photography, a real person looked into the lens, and that exchange of gazes formed the ethical foundation of the portrait. The eyes of an AI portrait have never looked at anyone. And yet we are moved by them — Ritchin calls this a "highly flawed but potentially interesting" simulation of humanity.³⁷

7. Bate's "Hands" Experiment

In a revealing practical experiment, Bate entered the same English prompt — "create a picture of two hands" — into ChatGPT and Ideogram, comparing the results.³⁸

ChatGPT produced an image referencing Michelangelo's Sistine Chapel ceiling (1510), translating the generic gesture into a stock-photo style. When asked for a non-religious version, it offered a secular "fist bump." Ideogram, by contrast, generated multiple images with diverse ages, cultures, genders, and skin tones, though with characteristic AI errors (incorrect finger counts, gratuitous symbolic elements like lavender flowers).

Bate draws three conclusions. First, different systems activate different latent archives, producing dramatically different results from identical prompts. Second, all results look "photographic" but periodically violate what Michel Foucault called the "discursive regularities" of photographic realism.³⁹ Third, and most consequentially, the responsibility for meaning shifts to the user — not as "creator" of images but as designer of their compositional meaning, representational ethics, and aesthetic effects.⁴⁰

8. What This Means for Us

Bate's concept of "composite memory" synthesizes everything above. AI-generated images are algorithmically condensed composites of existing cultural archives, manifested as artificial memories that never previously existed. This concept draws on the structural parallels between Galton's composite photography (1870s), Freud's dreamwork condensation (1900), and AI's latent-space computation (2020s).⁴¹

As Yacavone concludes, even though composite images lack indexical truth, they can — through artistic and activist intervention — enter the archive ecosystem as traces of "cultural reference memory." The very "a-historicity" of AI images may paradoxically become a historical mark of our era, testifying to the anarchival properties of the digital transition for future researchers.⁴²

Bate's final sentence deserves to be read slowly: "If humans leave 'photography' behind (the logic of photography as an aspect of subjectivity), they encounter something else: the machine-work images of algorithmic culture. In whatever practice, it is what humans see that matters to social forms of existence."⁴³

For those of us who work with Midjourney, Flux, DALL-E, or any generative image tool: the prompt is not a search query. It is an act of memory design — a conscious decision about which cultural memories to summon from the latent space, which to exclude, and what new composite memories to bring into existence. The theoretical frameworks assembled here — from Derrida's psychic archive through Freud's condensation to Steyerl's mean images — are not academic luxuries. They are the operating system for a practice that aspires to be more than the production of statistically averaged clichés.

---

Notes

1. Allison Parshall, "How This AI Image Won a Major Photography Competition," Scientific American, April 21, 2023.

2. David Bate, "AI Photography and Composite Memory," photographies 19, no. 1 (2026): 125–144.

3. Roland Barthes, Camera Lucida: Reflections on Photography, trans. Richard Howard (New York: Hill and Wang, 1981). Originally published as La chambre claire: Note sur la photographie (Paris: Cahiers du Cinéma / Gallimard / Seuil, 1980).

4. Kathrin Yacavone, "Virtual Photographs, Possible Memories: AI Images, the Archive, and the Works of August Sander and Elena Efeoglou," photographies 19, no. 1 (2026): 59–79.

5. Fred Ritchin, The Synthetic Eye: Photography Transformed in the Age of AI (London: Thames and Hudson, 2025).

6. Yacavone, "Virtual Photographs, Possible Memories," 59.

7. Jacques Derrida and Gerhard Richter, Copy, Archive, Signature: A Conversation on Photography (Stanford, CA: Stanford University Press, 2010); cited in Bate, "AI Photography and Composite Memory," 128.

8. Jacques Derrida, "Freud and the Scene of Writing," in Writing and Difference, trans. Alan Bass (London: Routledge, 1978).

9. Bate, "AI Photography and Composite Memory," 129. Also Jean Laplanche and Jean-Bertrand Pontalis, The Language of Psycho-Analysis, trans. Donald Nicholson-Smith (London: Karnac, 1988).

10. Bate, "AI Photography and Composite Memory," 129.

11. Sigmund Freud, The Interpretation of Dreams, trans. and ed. James Strachey, vols. 4–5 of The Standard Edition of the Complete Psychological Works of Sigmund Freud (London: Hogarth Press, 1953). Originally published 1900.

12. Bate, "AI Photography and Composite Memory," 130.

13. Bate, "AI Photography and Composite Memory," 130–131.

14. Francis Galton, "Composite Portraits," Nature 18 (1878): 97–100.

15. Galton, "Composite Portraits," 98. Cited in Bate, "AI Photography and Composite Memory," 131.

16. Allan Sekula, "The Body and the Archive," October 39 (1986): 3–64.

17. Bate, "AI Photography and Composite Memory," 131.

18. Catherine Malabou, Morphing Intelligence (New York: Columbia University Press, 2019). Cited in Bate, "AI Photography and Composite Memory," 132.

19. Freud, The Interpretation of Dreams; cited in Bate, "AI Photography and Composite Memory," 132.

20. Bate, "AI Photography and Composite Memory," 133.

21. Aleida Assmann, "Canon and Archive," in Cultural Memory Studies: An International and Interdisciplinary Handbook, ed. Astrid Erll and Ansgar Nünning (Berlin: de Gruyter, 2008), 97–108.

22. Roland Meyer, "The New Value of the Archive: AI Image Generation and the Visual Economy of 'Style'," IMAGE: The Interdisciplinary Journal of Image Sciences 37, no. 1 (2023): 100–111.

23. Hito Steyerl, "Mean Images," New Left Review 140/141 (2023): 82–97.

24. Daniel Palmer and Katrina Sluis, "The Automation of Style: Seeing Photographically in Generative AI," Media Theory 8, no. 1 (2024): 159–184.

25. Bernard Stiegler, Automatic Society, Volume 1: The Future of Work, trans. Daniel Ross (Cambridge, UK: Polity, 2016); cited in Bate, "AI Photography and Composite Memory," 134.

26. Parshall, "How This AI Image Won a Major Photography Competition."

27. Marianne Hirsch, Family Frames: Photography, Narrative and Postmemory (Cambridge, MA: Harvard University Press, 1997).

28. Bate, "AI Photography and Composite Memory," 135.

29. Jun Ahn, Until You Left Off Dreaming About (Tokyo: Case Publishing, 2024). Cited in Bate, "AI Photography and Composite Memory," 136.

30. Bate, "AI Photography and Composite Memory," 136.

31. August Sander, Citizens of the Twentieth Century: Portrait Photographs, 1892–1952, ed. Gunther Sander, text by Ulrich Keller, trans. Linda Keller (Cambridge, MA: MIT Press, 1986). Also Elena Efeoglou et al., "Elena Efeoglou Interviewed," in Artist Meets Archive #4 (Cologne: Internationale Photoszene Köln, 2025).

32. Yacavone, "Virtual Photographs, Possible Memories," 61.

33. Svea Braeunert, "From Evidence to Testimony: How AI-Generated Images of Refugees Can Make Demands," photographies 19, no. 1 (2026): 103–124.

34. Braeunert, "From Evidence to Testimony," 112.

35. Ariella Azoulay, The Civil Contract of Photography (New York: Zone Books, 2008).

36. Braeunert, "From Evidence to Testimony," 115.

37. Sara Oscar and Cherine Fahd, "Looking at Eyes That Never Looked Back: Photographic Portraits, Synthetic Images and (Mis)recognition," photographies 19, no. 1 (2026): 145–166.

38. Bate, "AI Photography and Composite Memory," 140.

39. Michel Foucault, The Archaeology of Knowledge (London: Tavistock, 1985). Cited in Bate, "AI Photography and Composite Memory," 140.

40. Bate, "AI Photography and Composite Memory," 141.

41. Bate, "AI Photography and Composite Memory," 142.

42. Yacavone, "Virtual Photographs, Possible Memories," 75.

43. Bate, "AI Photography and Composite Memory," 143.

What Is a Composite Memory?

A photograph records what was there. A composite memory reconstructs what was felt.

I use this term to describe the images I make — scenes built through AI that have no origin in any single moment, yet feel as though they belong to a past that was lived. They are not illustrations of memory. They are synthetic constructions that behave like memory: partial, atmospheric, emotionally weighted, and impossible to verify.

In cognitive science, memory is not a recording. It is an act of reconstruction — assembled each time from scattered fragments, shaped by present emotion, altered with every recall. What we remember is never what happened. It is what remains after everything else has fallen away.

This is how I approach image-making. I do not begin with a scene I want to depict. I begin with a residue — a sense of light in a space I may never have entered, the weight of stone in a building that does not exist, the stillness of water in a room where no one remains. The AI becomes a collaborator in this reconstruction, generating possibilities I could not have imagined alone, while I select and refine based on emotional recognition: does this feel like something I have known?

The result is an image that belongs to no archive, no location, no date. And yet it carries presence — the kind that makes you pause, not because you recognize the place, but because you recognize the feeling.

This is what I mean by composite memory. Not nostalgia. Not fiction. Something in between — where the boundary between experienced and imagined dissolves, and what remains is the emotional truth of a moment that may never have occurred.

— Avocado

Beginning

This site is a starting point, not a finished statement.

---

I work at the intersection of photography and AI-generated imagery — constructing scenes that function as composite memories. Images that never existed as photographs, yet carry the weight of something once felt.

Spaces That Remember is the first series presented here. More will follow as the work develops.

If you'd like to follow the process, this blog will document the thinking behind the images — how and why they are made.

— Avocado