Words and Pictures

1. The word that closes the conversation

One word appears repeatedly in discussions of generative artificial intelligence: fake. It is an efficient word. It is short, emotionally charged and difficult to answer without sounding defensive. It can describe a forged painting, a counterfeit handbag, an invented quotation, an impersonated voice, a fabricated news photograph or an image of an object that never existed. Once these things have been placed beneath the same label, their differences begin to disappear. Artificial construction becomes indistinguishable from fraud. Invention becomes evidence of dishonesty. The synthetic image is declared guilty before anyone has asked what it claims to be.

This is too blunt to be useful. A forgery is deceptive because it conceals its authorship and falsely claims another origin. A deepfake may be deceptive because it places words or actions into the body of a recognisable person. A fabricated documentary image is deceptive because it presents an invented event as something witnessed. These examples involve more than artificiality. They involve a false claim about identity, origin or occurrence.

An openly synthetic artwork makes a different claim. It may depict a person who was never born, a vessel that was never fired, a flower that never grew or a room that was never entered. Its subject is fictional, but fiction is not automatically fraudulent. The ethical question is not simply whether the represented thing existed. It is whether the image lies about what it is asking the viewer to believe.

The distinction I want to make is simple: simulation is not deception. Simulation constructs an appearance. Deception misrepresents the status of that appearance. The difference lies not only inside the image, but in the agreement surrounding it.

2. Images have always made agreements

No image is the thing it depicts. A painted pear is not a pear. It is pigment arranged so that colour, contour, shadow and highlight begin to behave like skin, weight and ripeness. A portrait is not a person. A landscape is not weather, distance or land. A photograph may carry a physical trace of something that stood before a lens, but it remains a selection: framed, timed, exposed, cropped and presented under particular conditions.

Images have always asked viewers to enter agreements. Painting asks us to accept that coloured matter can become flesh, cloth, metal, sky or light. Theatre asks us to believe temporarily in a room that is a set, a death that is a performance and a character who disappears when the actor leaves the stage. Cinema constructs continuity from fragments. Advertising transforms lighting, styling and retouching into desire. A museum display turns placement, typography and silence into authority.

None of this is necessarily dishonest. The viewer understands the terms, even when those terms remain partly unspoken. We know that an actor has not truly died. We know that the painted apple cannot be eaten. We know that a fashion photograph is an organised event rather than an innocent encounter with reality. The illusion is not a betrayal of the artwork. It is often its operating principle.

Generative AI does not invent simulation. It intensifies it. It produces images that may inherit the surface of painting, the apparent weight of sculpture, the tactility of printmaking or the optical confidence of photography without passing through the physical processes associated with any of them. The synthetic image speaks in the visual languages of older media while belonging materially to none. That instability is not an unfortunate side effect. It is one of the medium's central conditions.

3. What changes when the world is absent

Something nevertheless changes when an image no longer begins with a visible object, body or event. Photography established a powerful cultural relationship between image and occurrence. However staged or manipulated a photograph might be, light had still travelled from something in the world towards a photosensitive surface or digital sensor. The photograph therefore carried an index of presence: something had been there.

A synthograph breaks that chain. It does not record light from a scene. It generates the appearance of light. It may resemble a photograph, but no lens encountered its subject. A studio portrait can appear without a sitter, a studio or a camera. A still life can arrive without fruit, table, window or afternoon. The image possesses the visual consequences of an event that never took place.

This absence of origin gives synthography both its imaginative freedom and its ethical pressure. The image can go anywhere because it need not wait for the world to provide a subject. It can construct impossible materials, unavailable bodies, vanished architectures or entirely invented objects. But the same freedom allows an image to borrow the evidential authority of photography without possessing photography's connection to an occurrence.

That authority matters. A convincing shadow, an imperfect surface, a plausible lens effect or the restrained neutrality of institutional photography may imply observation even when no observation occurred. The eye recognises these signs and begins to build belief around them. As I have argued elsewhere, synthetic images persuade not because truth resides within them, but because they present familiar signals through which visual truth has historically been recognised.

This does not mean that realism itself is a lie. It means realism is a language whose authority can be used honestly or dishonestly. The existence of a knife does not determine whether it will carve wood or cause injury. The existence of a convincing image does not determine what claim will be made with it. Context, intention and presentation complete the act.

4. Simulation and deception

A simulation models the appearance or behaviour of something. A flight simulator is not a counterfeit aircraft. An architectural visualisation is not a fraudulent building merely because the walls have not yet been constructed. A medical model is not dishonest because it is not a living organ. A theatrical moon is not deceptive because it is painted canvas rather than a celestial body.

The simulation becomes deceptive when it is deliberately presented under false conditions: when the architectural visualisation is offered as a photograph of completed construction, when fictional footage is described as documentary evidence, or when an invented person is used to impersonate somebody real. Deception is therefore relational. It occurs between the artefact, the claim made for it and the audience receiving it.

A synthetic painting is not deceptive merely because no paint exists. It becomes deceptive if it is knowingly sold as a photographed physical canvas made by a named painter. A fictional portrait is not deceptive because its sitter never lived. It becomes deceptive if it is attached to a real identity or presented as evidence of an encounter. A generated ceramic vessel is not deceptive because clay was never touched. It becomes deceptive if the image is used to sell an object that does not exist or to fabricate the history of one that does.

The synthetic condition alone does not settle the ethical question. This is why the word fake often obscures more than it reveals. It treats material absence as though it were automatically moral concealment. It confuses the fictional subject with the dishonest artwork.

But art has always made things that do not exist. Mythological figures, imagined landscapes, impossible architectures, ideal bodies, invented histories and surreal encounters belong to the long territory of representation. We do not accuse a painter of fraud because a painted angel failed to attend the sitting.

The synthetic image extends this territory under new technical conditions. Its capacity for photographic plausibility makes disclosure more important, but plausibility does not cancel imagination. The fact that a fiction can look convincing does not disqualify it from being art. It asks us to become more precise about what kind of truth an image offers.

Simulation constructs an appearance. Deception misrepresents the status of that appearance.

5. Honest fiction

Fiction can be honest. An artwork may invent its subject while remaining truthful about its own status. It can say, explicitly or through context: this scene was constructed; this person is fictional; this object has no material counterpart; this image did not originate in front of a camera. Honesty does not require the artwork to abandon illusion. It requires the illusion not to be falsely presented as evidence.

In my own practice, the term synthography is part of that honesty. It names the process rather than hiding it. I use generative AI to make images that borrow the appearances of photography, painting, drawing, printmaking, sculpture, craft and design. The images may simulate material histories they never possessed, but they are presented as synthetic artworks. They do not claim that a brush touched the surface, that a kiln was fired or that a camera witnessed the scene. The object is absent. The construction is acknowledged.

In Simulacra, fruit and flowers appear to have been painted as small oil studies. The images carry scumbled grounds, dragged brushwork, modest arrangements and the memory of an attentive hand. Yet no fruit was placed, no flower cut, no board prepared and no paint moved. The project does not conceal this absence; the absence is its subject. The works ask what remains of painting when its material process has been replaced by a synthetic memory of that process. That is simulation used as enquiry rather than disguise.

FEED ME approaches the matter differently. It borrows the absurd visual grammar of AI slop — the dancing animal, the miraculous object, the manufactured spectacle — but subjects it to structure, refinement and satire. It does not deny that synthetic imagery can be empty, manipulative or produced as waste. It distinguishes automation from authorship, speed from intention and appetite from meaning.

Neither project needs to pretend that the images are materially what they resemble. Their honesty lies partly in refusing that pretence. This does not make them less illusory. It allows illusion to become visible as a deliberate artistic material.

6. Authorship without the solitary hand

The question of honesty also extends to authorship. A synthographer does not execute an image mark by mark in the manner of a draughtsman, painter or illustrator. The system generates the visible form. Pretending otherwise would merely replace one confusion with another. But the absence of direct execution does not mean the absence of authorship.

A synthographic image begins with conditions: an idea, a subject, a visual language, a set of constraints, a relationship between images and a reason for making them. Language initiates the process. The model produces possibilities. The artist evaluates what has appeared.

An image is accepted, rejected, revised, redirected or replaced. A project develops rules. Motifs recur. Others are prohibited. The sequence acquires rhythm. Accidents are recognised as either useful or empty. The artist decides when an image is merely fluent and when it has become necessary.

I have previously described the synthographic image as being authored into being rather than taken from the world. Its construction depends upon intention, language, structure and constraint, followed by selection, refinement and refusal. This authorship is not identical to traditional craft, and it gains nothing by pretending to be. It belongs to direction, negotiation and judgement. The machine generates possibilities. The artist decides which possibility belongs to the work.

That decision may happen hundreds of times across the making of a single project. Authorship gathers not only around the isolated image but around the complete structure: the title, the concept, the visual constitution, the transformations across a sequence, the writing, the order, the exclusions and the manner in which the work is presented.

A generator can produce an attractive picture quickly. It cannot independently decide why that picture should exist, what intellectual pressure it carries, what it must refuse or how it belongs within a larger practice. The image may be generated. The work is composed.

7. Responsibility without apology

To argue that simulation is not deception is not to argue that synthetic image-making is harmless in every context. The greater the visual fluency of generative systems becomes, the greater the responsibility attached to their use. A photorealistic image can misidentify people, invent events, fabricate evidence, manipulate emotion and exploit the viewer's inherited trust in photography. Disclosure matters most where confusion is likely to cause harm or where an image is presented as information rather than imagination.

There are also wider questions concerning training data, labour, consent, representation, environmental cost, aesthetic homogenisation and the concentration of technological power. A serious artist working with generative AI cannot reasonably behave as though the medium exists outside those conditions.

But seriousness does not require permanent apology, nor does responsibility mean allowing the worst uses of a technology to define every possible use. Photography can document atrocity, sell perfume, falsify appearances, preserve family history, produce surveillance or become art. The camera is not ethically resolved by its mechanism. Meaning emerges through use, context and consequence. The same is true of generative image-making.

Some synthetic images are deceptive. Some are exploitative. Some are repetitive, generic or empty. Some are created as industrial quantities of disposable content. That reality should be acknowledged without turning it into a definition of the medium itself.

In factual work, machine hallucination can be dangerous because invention is mistaken for knowledge. In artistic work, invention may be the material from which meaning is made. The difference lies in how the invention is disciplined, contextualised and understood. As I wrote in The Tripping Machine, an undisciplined hallucination becomes slop; a disciplined hallucination may become art. The artist's task is not to prove that every synthetic image deserves defence. It is to make distinctions visible through practice.

That means being honest about the tools, careful about the claims, critical of the outputs and unwilling to confuse abundance with achievement. It means understanding that technical polish is not meaning, that plausibility is not truth and that novelty is not originality. It also means allowing the work to be beautiful, comic, erotic, decorative, unsettling or quiet without forcing every image to carry a manifesto on its back. The backlash can sharpen the questions. It should not be permitted to choose all the answers.

8. The object is fictional; the artwork is real

A synthetic image may show us something that never stood in the world. Its ceramic surface may never have cooled after firing. Its graphite line may never have pressed into paper. Its flower may possess no botanical origin. Its body may have no childhood, memory or pulse. Its shadow may describe an event that never passed through time.

These absences are real, but so are the decisions that give the work form: the conception, writing, direction, selection, refusal, sequencing and presentation. So is the encounter between the viewer and the image. So are the questions the work opens concerning touch, evidence, tradition, belief, desire and material presence. The object may be fictional. The artwork is not.

To call a synthetic image fake simply because its subject did not exist is to mistake representation for evidence and fiction for fraud. It is to demand that all images justify themselves through material origin, as though imagination were a defect and simulation a confession.

I would rather ask more exact questions. What is the image claiming? What does it disclose? What kind of belief does it invite? Whose authority does it borrow? What responsibility accompanies its realism? What has the artist shaped, and what has been left to automation? Does the work conceal its artificiality, or use artificiality as its material?

Honesty is not the enemy of illusion. It is what allows illusion to operate without becoming deceit. A synthograph does not need to apologise for lacking a camera, a sitter, a kiln, a brush or a physical object. It needs to understand the significance of those absences and present them without falsehood.

Simulation constructs an appearance. Art gives that appearance purpose. Deception begins only when the appearance is made to lie about what it is.

Simulation is not deception. It is a way of making fiction visible — and, when handled honestly, a way of making truth newly difficult, strange and alive.

1. The sentence that opened the door

I recently found myself thinking about a sentence: I am a synthographer negotiating with the ghost of painting. It arrived while looking at an image of a male figure rendered in a style that seemed to hover between studio portrait, life study and Abstract Expressionist surface. The image was not a painting, of course. No brush had touched it. No body had stood in front of a canvas. Yet painting was unmistakably present, not as a medium, but as a memory.

That is where the thought began to widen. A synthographer is not only negotiating with the ghost of painting. The negotiation is also with photography, sculpture, printmaking, product design, fashion imagery, museum catalogues, advertising, cinema, stock photography and the countless visual systems through which the world has already been organised. Generative AI does not arrive in an empty room. It arrives in a room crowded with ghosts. This is not a frightening idea, but it is a strange one. The synthetic image is new, but it is never innocent. It carries the atmosphere of older media because it has been trained on their traces. It speaks in inherited accents. It borrows gestures, surfaces, poses, lighting, compositions and signs of authenticity. The question is not whether these ghosts are present. The question is how consciously we choose to work with them.

2. The training set as haunted archive

Modern image generators are built from vast collections of images and text. Their training sets contain photographs, drawings, paintings, screenshots, diagrams, illustrations, product images, fashion editorials, interiors, landscapes, faces, bodies, objects and captions: an enormous visual weather system gathered from the web. To call this material "data" is accurate, but it also makes it sound cleaner than it is. Data is never just data. It carries habits.

A photograph of a painting is already a translation. The original object has passed through a camera, a lens, a file format, compression, colour correction, metadata, search terms, captions and online display. A sculpture in a museum catalogue has become an image of an object, flattened into light. A body in a fashion campaign has already been styled, lit, directed, retouched and circulated. By the time such images enter a model, they are not raw reality. They are cultural artefacts.

This is why I think of the training set as a haunted archive. The model does not remember like a person, but it is haunted like a house. It does not recall a painting, a photograph or a statue in any human sense. Instead, it carries statistical residues of visual experience: the probability of a shadow, the expectation of a pose, the look of oil paint, the shine of ceramic, the authority of a white background, the persuasive calm of studio light. The generated image emerges from this residue. It is not copied from one source, but assembled from an atmosphere of sources. It is born from accumulated conventions.

3. Borrowed authority

This gives synthography its peculiar power. An AI-generated image can borrow the authority of photography without being a photograph. It can borrow the surface of painting without being painted. It can borrow the mass of sculpture without occupying space. It can borrow the grain of printmaking without passing through a press. These borrowings are not mistakes. They are part of the medium's grammar.

Photography gives the synthetic image its first great disguise: plausibility. A generated body can appear to have stood under real light. A fabricated object can cast a convincing shadow. A room can seem inhabited by dust, air and time. The camera is absent, but its authority remains. We are trained to believe certain optical cues because photography spent nearly two centuries teaching us how truth might look.

Painting offers something different: surface, gesture and historical depth. When a synthetic image appears painterly, it summons not only brushwork but the long cultural romance of touch. We see marks and imagine a hand. We see colour and imagine pigment. Even when no paint exists, painting lends the image a kind of emotional weather. Sculpture contributes objecthood. It lets the artificial thing feel made, weighted, placed. Printmaking contributes edges, pressure, registration, accident and repetition. Design contributes order, clarity, display and seduction. Synthography moves through all these older languages, borrowing from each while belonging fully to none. That "belonging to none" matters. It is not a weakness. It is the new territory.

The machine generates possibilities. The artist decides which apparition matters.

4. The synthographer as negotiator

If the model is haunted, the synthographer becomes a negotiator. The work is not simply to write a prompt and receive an answer. It is to recognise what has appeared, what has failed, what has become interesting by accident, and what must be refused. The process is part instruction, part correction, part listening.

A prompt begins as an intention, but the model answers through its own inherited visual logic. Ask for a ceramic vessel and it may produce museum lighting, auction-house neutrality, Japanese glaze, studio photography, catalogue space and the ghost of craft. Ask for a male figure and it may return fitness imagery, classical statuary, fashion editorial, anatomical study, portraiture and desire. The machine does not know these things as meanings, but it has absorbed their appearances as patterns. The synthographer works inside that tension, steering the model towards a subject while guiding it away from cliché, asking for material presence while knowing there is no material, and constructing images of objects that were never built, bodies that were never photographed and flowers that never grew.

The work happens in the negotiation between language, model, visual memory and selection. This is why authorship in synthography feels different, but not absent. The artist is not the sole maker in the old sense, but neither is the machine. The image is produced through collaboration with a system trained on the visual past. Working well with it requires taste, patience, suspicion and recognition. Especially recognition. The machine generates possibilities. The artist decides which apparition matters.

5. The ghost is the point

It is tempting to describe the ghostliness of AI-generated images as a problem: evidence of imitation, derivation or unreality. I think the opposite may be more useful. The ghost is the point. Synthography is not pure invention from nowhere. It is invention after everything.

Every medium carries the memory of earlier media. Photography inherited painting's compositions. Cinema inherited theatre, photography and literature. Digital design inherited print. Generative AI inherits almost all of them at once, not in a neat historical line, but as a dense cloud of visual influence. It is a medium built from the afterimages of other mediums. This does not make the synthetic image less interesting. It makes it more unstable, and therefore more alive. The image looks back at us with borrowed eyes. It asks us to recognise the systems that taught us how to see: the studio, the museum, the catalogue, the screen, the advertisement, the archive. It asks what remains of photography when no camera was present, what remains of painting when no hand touched the surface, what remains of sculpture when there is no object to walk around.

A synthographer negotiating with ghosts is not trying to exorcise them. The task is not to remove painting, photography, sculpture or print from the synthetic image. The task is to understand how they persist inside it, how they change when summoned by a machine, and how their old authority can be turned into something newly strange. The haunted image is not empty. It is crowded. It is made from traces, echoes, refusals and returns. It belongs to no single medium because it has passed through too many. Perhaps that is where synthography begins: not with the invention of a new image, but with the recognition that every image is already inhabited.

1. No machine elves

When we say an AI model "hallucinates", we are not saying it is conscious, intoxicated, spiritually expanded, mentally ill or having a private experience. It has no inner theatre. No pink elephants. No cosmic insects. No Terence McKenna machine elves sitting behind the graphics card eating luminous fruit.

The term is metaphorical, and not an entirely settled one. In artificial intelligence, hallucination usually describes output that is unsupported, factually incorrect or inconsistent with the information a system has been given. The exact definition varies between tasks and fields, which is worth remembering before we make the machine sound as though it has dropped acid.

Human hallucination is something very different: a perception-like experience occurring without the corresponding external stimulus. A voice may be heard without a speaker, or a figure perceived where no body is present. Such experiences can arise under many different conditions. The important point here is simply that something generated within perception can be experienced as though it came from outside.

That is why the metaphor has survived in AI. Generative systems also produce convincing appearances without secure attachment to reality. They make things up. Sometimes they do so badly, producing errors that are easy to spot. Sometimes they do so beautifully, producing fictions that appear calm, coherent and almost inevitable.

The resemblance ends there, but the metaphor remains useful. The machine is not having a trip. The image sometimes looks as though it is.

2. When hallucination is failure

In a language model, hallucination is normally a problem. The system invents a quotation, fabricates a source, misnames a book, supplies an incorrect date or answers with great confidence from a place of no reliable knowledge. This is not creativity in any useful sense. It is falsehood wearing a good suit.

The danger lies partly in fluency. A fabricated answer can be grammatically polished, tonally assured and completely persuasive. The surface carries authority even when the structure underneath it does not. For a system being used to retrieve or communicate facts, that matters enormously.

Image generation complicates the word. A generated image may reasonably be described as hallucinatory when it contradicts a prompt, invents requested details incorrectly or produces relationships that violate ordinary spatial or physical logic. But an imaginary glass flower is not automatically a hallucination in that technical sense merely because no such flower exists. If I asked for an imaginary glass flower, invention was the assignment.

That distinction matters enormously to synthography. The synthetic image is allowed to show things that never stood before a camera. A fictional portrait is not factually incorrect simply because its sitter was never born. A cabinet does not become an error because no carpenter made it. In artistic image-making, invention is not necessarily the malfunction. It may be the point.

3. The trip learns manners

Earlier image generators made the metaphor almost too easy. Extra fingers, melting faces, broken teeth, nonsense text, fused objects, impossible anatomy, waxy skin and dead eyes announced themselves like bad actors entering too early. Hands became spiders. Furniture grew limbs. Words dissolved into decorative insects. The world seemed to have been poured into the machine and stirred with a dirty spoon.

Those images could be beautiful, strange and occasionally exhilarating, but the instability was visible across the surface. The generator often seemed unable to decide where one object stopped and another began, how a hand connected to an arm, whether lettering was language or ornament, or whether a chair required four legs merely as a matter of convention.

That period has not disappeared completely, but the technology has changed quickly. Newer image systems follow instructions more closely, allow increasingly precise editing and can use reference images to preserve details across iterations. Generation is becoming more controllable: one element can be changed while another is retained, identities can persist, references can be combined and an existing image can be refined rather than abandoned and started again. The trip has, to some extent, been tamed, but it has not been abolished.

High visual fidelity does not necessarily mean that the depicted world makes sense. This makes contemporary failures more interesting because the hallucination has learned manners. It now arrives dressed as plausibility.

A handle may look beautifully made but be impossible to pull. A hinge may be mounted where no door could open. A clasp may possess no function. A strip of timber can appear convincingly attached while having no believable means of attachment. The image seems reasonable until someone asks it a practical question.

4. The imaginary carpenter

This became especially clear to me while developing Arca II, a project about fictional cabinets made from reclaimed timber, old floorboards, shed wood, sign fragments, rusted hinges, peeling paint and occasional pieces of driftwood.

The work developed through conversation, generation, looking, correction and return. Rather than treating the first successful-looking image as finished, I began interrogating the object inside it. Could that door actually open? Why is the latch there? What is this piece of timber attached to? Would the cabinet stand? Would a carpenter really make that joint?

These questions altered my relationship with the machine. I was no longer simply asking whether the picture looked attractive. I was asking whether the hallucinated object could survive its own visual logic.

A handle needed to behave like a handle. A latch needed a reason to exist. Hinges needed to connect things that could plausibly move. Timber had to feel joined rather than pasted onto the surface as decorative debris. This was not the elimination of invention. It was the disciplining of it.

Working this way began to feel less like ordering an image and more like sitting beside an imaginary carpenter at a bench. The carpenter was brilliant, tireless and slightly deranged. It could produce a convincing cabinet from nothing, but it might also attach a beautiful brass fitting to a place where the fitting could perform no earthly function. My job was not to sober the carpenter up completely. It was to keep the good ideas and confiscate the power tools when necessary.

5. Set and setting

This is where the psychedelic analogy becomes useful again, provided we continue not to take it literally. The phrase set and setting, closely associated with Timothy Leary and the psychedelic culture of the 1960s, describes the importance of expectation and surrounding conditions in shaping a psychedelic experience. The history is more complicated than one man or one slogan, but the basic idea is useful here.

Generative image-making has its own peculiar version of set and setting. There is the prompt, the chosen model, the reference image, the conversation around the image, the corrections, the visual constraints, the crop, the sequence and the willingness to reject an almost-good result. There is also the artist's taste: everything already learnt about bodies, materials, photography, painting, design, construction and the difference between an interesting accident and an irritating one.

These conditions shape what is eventually permitted to emerge. The machine may be doing the generating, but it is not working in an artistic vacuum. The synthographer becomes a kind of trip-sitter. Sometimes the task is to encourage the strange turn rather than suppress it. Sometimes it is to calm the excess. Sometimes the vision needs redirecting because the machine has mistaken ornament for structure, detail for meaning or plausibility for truth. And sometimes the correct response is simply: absolutely not, you've given him seven fingers again.

6. Disciplined hallucination

This is why I do not want to understand hallucination only as a bug. Technically, the word has legitimate uses for describing generative failures. In factual systems those failures can be serious because unsupported invention is being presented where knowledge is expected. But artistic invention operates under a different agreement.

When I use hallucination to describe synthography more broadly, I am using it deliberately as a metaphor for the generative condition: language, references and learned visual relationships enter a probabilistic system, and an appearance emerges for something that may never have existed anywhere else.

The distinction is important. The machine has not hallucinated a fictional cabinet merely because the cabinet is fictional. The hallucinatory quality lies in the peculiar way the image can summon an appearance without first requiring the world to provide the thing.

From an artistic perspective, the question therefore becomes not whether invention occurs, but what happens to it afterwards. An undisciplined hallucination becomes slop. A disciplined hallucination may become art.

That discipline is not the enemy of surprise. I would not want every strange turn removed from generative image-making. Some of the medium's most interesting possibilities appear precisely because the system has not followed the expected route. It associates strangely. It exaggerates. It invents solutions no sensible craftsperson would propose. Sometimes those solutions are ridiculous; sometimes they are the reason to keep looking.

The artist's role is not to make the system entirely obedient. A perfectly obedient machine might become rather dull. The role is to recognise when deviation produces possibility and when it merely produces nonsense.

7. What comes back from the trip

The more serious problem now is not that AI-generated images always look obviously strange. That was the easy phase.

A synthetic cabinet may appear handmade. A fictional portrait may seem observed. A generated flower may carry the authority of botanical study. A scene that never happened may present the visual signals of something calmly photographed on an ordinary afternoon.

The trip has become quieter, sharper and more persuasive. It no longer always shows us monsters. Sometimes it shows us an ordinary object and asks us to believe in it.

Synthography lives inside that tension. It does not record the world, but it borrows the world's behaviours. It does not require physical presence, but it can reproduce many of the signs through which presence is ordinarily recognised.

Hallucination, then, provides both a technical warning and a useful artistic metaphor. The two meanings should not be confused, but neither needs to cancel the other. One tells us where generative systems fail. The other gives us a way of thinking about their astonishing capacity to invent appearances.

My task is to work somewhere between them: to look harder, refuse more often, correct with care and leave enough room for the machine to surprise me. Every convincing synthograph is, in this sense, a negotiated hallucination. The machine is still tripping. The artist decides what comes back from the trip.

1. The eye wants coherence

A synthetic image can show us something that has never existed and still make it feel entirely plausible. A glass flower catches light along its edges. A ceramic vessel carries the small irregularities of firing and handling. A forged metal chrysanthemum seems heavy enough to lift from the table. A fictional portrait appears to have been made beneath studio lights by a camera positioned a few feet from the sitter. None of these things needs to have existed outside the image. Yet the eye may accept them almost immediately.

This is the problem that interests me in synthography. If an image does not have to originate in an actual object, body or event, why can it still feel so convincing? What exactly are we responding to when we look at something impossible and think, without much hesitation, yes, that looks right?

The answer cannot simply be realism. Realism is an effect, not an explanation. Nor can it simply be detail. Generative systems are capable of extraordinary amounts of detail, but an image can be intricate and still feel completely wrong. What persuades us is something quieter. The parts agree with one another.

2. Seeing is an inference

Vision does not give us direct access to the physical properties of things. Light reaches the eye, and from the resulting patterns we infer shape, depth, surface, material, distance and illumination.

Material perception makes this especially clear. Gloss, for example, is not contained in a single bright highlight. Our judgement depends upon relationships between reflections, surface shape, viewpoint and lighting. Experiments have shown that people can misperceive gloss when these relationships change even though the physical reflectance of the depicted material has not. Expectations about what an object is made from can also influence how its material qualities are perceived.

Seeing therefore involves interpretation. I do not mean that every glance is a conscious calculation. Quite the opposite. Much of the work happens so efficiently that the conclusion simply arrives. A surface seems glossy. A cloth seems soft. A shadow tells us that an object sits on a table rather than floating above it. Perspective suggests depth. Occlusion tells us which form is in front. A highlight follows a curve and helps us understand both the curve and the material covering it.

These cues do not prove that an object exists. They help us understand what kind of object we appear to be seeing.

A synthetic image enters this perceptual system carrying exactly the same kinds of visible relationships. The eye has no special category of light reserved for fictional objects. It responds to what is present in the image.

3. Material behaviour

This is why I increasingly think about behaviour rather than detail when making synthetic images. If something is meant to look like glass, simply making it transparent is not enough. Glass refracts, reflects and transmits light. Its thickness affects the appearance of its edges. A curved glass surface distorts what can be seen through it.

Metal behaves differently. A polished surface returns its surroundings sharply; a rough or oxidised one scatters light differently and interrupts the reflection. Fabric folds according to gravity, thickness and tension. Timber grain tends to follow the structure of the material rather than wandering arbitrarily across it.

Research into human material perception repeatedly shows that our judgements emerge from combinations of image information rather than from one perfect visual clue. Shape, illumination, reflectance and texture interact, and under some conditions those interactions can produce convincing perceptual errors.

That helps explain something I encounter constantly in my own work. A beautifully rendered synthetic object can fail because one small relationship is wrong. A shadow disagrees with the light. A metal edge behaves like plastic. A piece of wood has convincing grain but impossible construction. A flower looks delicate but its stem could not support its apparent weight. The failure is not always obvious enough to name immediately. The image simply stops persuading.

Conversely, an image does not need microscopic perfection if its larger relationships make sense. A modest amount of surface information, correctly organised, can carry more authority than spectacular detail without structure. Belief enters through coherence.

4. Photographic authority

Photography adds another layer because we have spent nearly two centuries becoming familiar with its visual conventions. Depth of field, lens perspective, exposure, motion blur, studio lighting, flash, grain, sensor noise and countless other characteristics help an image behave photographically. They do not establish that a photograph was actually taken, but they contribute to the appearance of photographic capture. Generative systems have become increasingly capable of reproducing these conventions.

The consequences are measurable. In a controlled 2022 study, participants were unable reliably to distinguish the particular AI-synthesised faces used in the experiment from photographs of real people; remarkably, the synthetic faces were also judged, on average, as more trustworthy. That result does not mean that every generated face fools every viewer, nor that visual realism is now perfect. It demonstrates something narrower and more interesting: under the right conditions, photographic plausibility can survive the complete absence of a photographed person.

That matters because photographic appearance once carried an especially strong implication of origin. As I explored in My Camera Never Lies, a photograph traditionally begins with an exposure to something present before the camera. A synthetic image can now reproduce many of the visible consequences of that encounter without the encounter itself. The image can contain the look of evidence without being evidence of the depicted event.

That distinction is essential. The synthetic portrait may persuade me that the light is plausible, the skin coherent, the expression human and the optical space photographic. None of those perceptual judgements establishes that the man existed. Visual credibility and factual origin are different things.

5. Persuasion is not proof

This is where the word truth in the title needs some care. I am not suggesting that a correct shadow is truthful, or that convincing skin somehow turns fiction into fact. The "signals of truth" are better understood as signals we have learned to associate with coherent appearances.

Cinema has always relied upon this distinction. A set does not need to be an actual nineteenth-century street in order for perspective, light, scale, costume and texture to establish a persuasive one on screen. Painting can make fruit feel heavy without giving us fruit. Architectural visualisation can make an unbuilt room intelligible before its walls exist.

Synthetic imagery extends that old capacity into a new technical condition. It can construct an entire visual proposition without first constructing the thing being depicted.

This is precisely why I want to keep this essay separate from the ethics of deception. Whether an image is labelled honestly, used responsibly or presented with the intention to mislead belongs to another discussion. Here the question comes earlier: What allows the image to work at all?

The answer seems to lie not in some single switch labelled realism, but in the accumulation of compatible cues. Light agrees with form. Form agrees with material. Perspective agrees with space. Detail appears at an appropriate scale. The surface carries enough irregularity to escape sterile perfection without becoming noisy.

When these relationships reinforce one another, the image becomes easier to accept perceptually. When they contradict one another, the spell weakens.

6. The signals of truth

My own work increasingly takes place at this threshold. I am less interested in making impossible things look spectacular than in giving invented things enough internal logic to hold their ground. Across projects involving ceramics, metal flowers, fabricated objects, male figures, drawings and photographic fictions, I keep returning to the same practical question: What would this thing have to do visually in order to feel as though it could exist?

A chrysanthemum made from blackened steel does not persuade because every petal is rendered at maximum resolution. It persuades when the petals appear to have been formed from metal, when the stem seems capable of supporting them, when the finish catches light appropriately and when small signs of making belong to the imagined process.

A portrait works similarly. Skin, posture, expression, focal behaviour, light and atmosphere support one another until the fictional sitter occupies the frame with enough consistency to feel present. This is not a search for flawless simulation. Flawlessness can be suspicious in its own way. What I want is coherence: enough agreement between the parts for the proposition to survive looking.

That is why the synthetic image is such an interesting test of visual belief. Remove the actual vase, flower, body or room and the eye does not suddenly stop working. It continues to examine relationships, recognise materials, infer depth and respond to familiar visual structures.

The world may be absent from the origin of the depicted scene, but knowledge of how the world tends to look remains essential to making the scene persuasive.

That, for me, is where the signals of truth reside. Not in proof. Not in factual history. In the small agreements through which an image becomes visually coherent enough to be believed on its own terms. Synthography does not show that truth is irrelevant. It shows something more precise: looking real and being real were never the same judgement.

1. These are not apples

Apples carry a surprising amount of cultural baggage. There is Snow White's poisoned apple, Cézanne's persistent return to apples in still life, the Big Apple and Apple the technology company. They are ordinary enough to recognise instantly and loaded enough to carry meanings almost without trying. Cézanne returned to them repeatedly, testing what such an unremarkable object could do inside a painting. That is why apples are useful here. Because these are not apples, and this is not a photograph.

What appears to be a glass bowl of fruit was not placed on a table and photographed. No apples reflected light towards a lens. There was no fruit bowl, studio or decisive moment waiting to be captured. The image was generated using artificial intelligence and developed through a process of instruction, selection and refinement. I call the practice synthography, and the resulting image a synthograph.

In the simplest terms, synthography is the intentional creation of images using generative artificial intelligence, in which significant visual content is synthesised by a computational model rather than optically captured from a scene or manually constructed mark by mark by the artist. That definition sounds simple. The interesting part begins when we ask what it actually means.

2. A word with a history

I did not invent the word synthography. Artist and academic Elke Reinhuber proposed the term in work presented at the ArtsIT conference in 2021 and published in 2022, arguing for a way of describing synthetically produced images beyond conventional photography. She returned to the subject in later writing, particularly in relation to photorealistic artificial images produced without optical capture.

Since then, synthography has appeared in academic discussion, photography education and the language used by artists and image-makers themselves. It is not the only term in circulation. Promptography, synthetic photography and other descriptions are also used, and no single vocabulary has yet become universal. That uncertainty does not make the word meaningless. It means the field is young and its language is still settling.

The construction of the word is useful. The first part, synth-, points to the synthetic: something produced through artificial combination or construction rather than simply found in nature. Synthetic itself ultimately belongs to the family of words derived from the Greek synthesis, a putting together or composition. The second part, -graphy, is familiar from words concerned with writing, drawing, representation and recording. Synthography therefore names an image-making practice based upon synthetic production. I like the term because it describes how the image comes into being without pretending that the machine is the artist.

3. What I mean by synthography

My own use of the term is broader than a definition restricted to photorealistic artificial photography. A synthograph may look photographic, but it does not have to. It can resemble a charcoal drawing, an oil painting, a screenprint, a botanical plate, a piece of sculpture, a textile sample or something that belongs comfortably to none of those categories.

What makes it synthographic is not its style. It is the generative condition under which substantial parts of the image are produced.

The artist begins with an intention and establishes conditions. Those conditions may be expressed through language, reference images, sketches, masks, previous generations, compositional constraints or combinations of them. A generative system interprets those inputs and produces visual possibilities. The artist responds: accepting, rejecting, redirecting, correcting, editing and trying again.

The machinery varies between systems and will continue to change. Text-to-image generation has been central to the recent development of the field, but contemporary synthographic practice already extends well beyond the idea of somebody typing one prompt and receiving one finished picture. That is why I would not define synthography as prompting.

A prompt can be part of synthography in the same way that pressing a shutter can be part of photography. Neither action, by itself, describes the whole practice.

4. Where the boundary lies

The presence of artificial intelligence does not automatically make an image a synthograph. Modern photography already contains enormous amounts of computation. Cameras reduce noise, sharpen images, interpret colour and may combine information from several exposures. Editing software can alter tone, remove distractions or reconstruct small areas. None of this necessarily changes the basic fact that a photograph began with light arriving from something in front of a camera. The distinction becomes clearer when we ask where the depicted content came from.

If I photograph an apple and use software to adjust its colour, I still have a photograph of an apple. If I ask a generative system to create an apple that was never there, something different has happened.

Between those examples lies an expanding hybrid territory. A photograph may contain generated extensions. A synthetic image may incorporate photographic references. A real face can be placed inside an invented setting. Generative editing can make the boundary difficult to locate. That does not invalidate the distinction. It means the boundary behaves more like a continuum than a wall.

For my purposes, the useful threshold is substantial generative production of the depicted content. When the scene, subject, object or visual form is principally being synthesised rather than recorded, I am comfortable calling the result synthographic.

5. Not photography, not painting

Synthography can borrow the appearance of almost any older medium, but appearance and process are not the same thing. A synthographic portrait may use the visual language of studio photography without having involved a sitter or photographic exposure. A synthetic charcoal drawing may contain pressure, dust and erasure without graphite touching paper. A fictional ceramic object can display glaze, cracks and fired irregularity without clay or heat.

This is why I do not think synthography is simply a strange subdivision of photography. Photography begins with an encounter between light and something present. Synthography does not require that encounter.

Nor is it digital painting under another name. A painter, whether working physically or digitally, normally constructs the image through direct acts of visual inscription: mark, stroke, shape, colour and erasure. The synthographer delegates a substantial part of that low-level visual execution to a generative system and works instead through conditions, responses and decisions. This does not mean the artist surrenders control. It means control operates differently. The image is not executed piece by piece in the traditional sense. It is negotiated into form.

6. Where authorship lives

This question of authorship matters because generative AI can produce an attractive image extremely quickly. That fact is sometimes treated as though it settles the argument. It does not. The ability to produce an image and the ability to make a body of work are different things.

In my own practice, authorship begins before generation. It starts with the question I want a project to ask, the visual territory it will occupy, the materials or behaviours I want to simulate and the rules I establish around colour, format, subject and sequence. Generation then produces possibilities. Most of them do not survive.

I reject images because the composition is wrong, because a material behaves badly, because an object makes no structural sense, because a face has drifted, because the image is too obvious, too decorative, too polished or simply because it does not belong with the others.

Authorship therefore accumulates through conception, instruction, iteration, selection, refusal, correction, editing, sequencing, writing and presentation. The process is exploratory and often recursive: an image changes the idea, the idea changes the next image, and the developing project gradually establishes standards that later generations must meet.

The generative system contributes enormously to the visible result. There is no reason to minimise that contribution. But contribution is not the same thing as intention. The artist provides the reason.

7. Why call it a medium?

There is a temptation to call everything made with generative artificial intelligence simply AI art. I understand the convenience, but I find the phrase increasingly unhelpful. Artificial intelligence is a family of technologies, not an aesthetic position. Saying AI art tells me something about the machinery involved but very little about the kind of work that has been made. Synthography is more specific. It names an image-making practice organised around synthetic generation.

Whether it ultimately becomes the dominant term is impossible to know. Vocabulary evolves with technology, and another word may eventually prove more useful. For now, however, synthography has an identifiable history, a growing cultural and scholarly usage and enough conceptual precision to describe a genuinely new condition of image-making. I am therefore comfortable calling it an emerging medium.

Not because every generated picture is art. Not because a new tool automatically deserves a new artistic category. And not because photography, painting, illustration or any other older medium has somehow become obsolete.

I use the word because the relationship between artist, image and world has changed enough to need describing. A photograph is produced through an encounter with light from something present. A painting or drawing is constructed through marks. A synthograph is generated from conditions established by the artist and shaped through judgement. That is the definition I find most useful.

Synthography is not photography without a camera, painting without paint or illustration without a hand. It may borrow from all of them, but it has its own peculiar logic: the artist defines, directs and judges; the system synthesises; the image emerges through the negotiation between them. The apples do not exist. The image does. That is where synthography begins.

1. Too many images

I sometimes wonder whether I have become allergic to images. This is an odd thing for an artist to admit, especially one whose practice depends almost entirely upon making them. But there are moments now when I look at a screen and everything seems to have collapsed into the same visual substance. Photographs, advertisements, infographics, memes, generated illustrations, promotional graphics, motivational quotations, diagrams, thumbnails, avatars, news images, product shots: they arrive one after another until the distinction between information and decoration begins to disappear.

I notice this particularly on social media. Even words increasingly arrive disguised as pictures. A paragraph becomes a coloured card. A statistic becomes an infographic. A thought becomes typography arranged over a gradient. Everything has to become visual before it is allowed to compete for attention. After a while, I stop reading. I simply see image.

Generative artificial intelligence has accelerated this condition enormously. An image that might once have required a photographer, illustrator, designer, studio or several hours of work can now be produced in seconds. The barriers have fallen, which is creatively exciting, but abundance has consequences. When almost anyone can make almost anything almost immediately, the problem is no longer access to images. The problem is deciding which images deserve to exist.

2. The arrival of slop

There is now a useful phrase for some of what has followed: AI slop. It is deliberately unflattering, but I think the term describes something real. Slop is not simply an AI-generated image that somebody dislikes, nor does it mean that generative tools inevitably produce bad work. It is what happens when production becomes detached from judgement: images made because they can be made, multiplied because multiplication is easy, circulated because circulation itself has become the objective.

Its defining feature is not ugliness. Some AI slop is extremely attractive, and that may be part of the problem. It glows beautifully. It sparkles. Every surface is polished, every sunset incandescent, every fantasy city enormous, every face implausibly symmetrical. There are intricately armoured warriors, luminous forests, impossible architecture, women surrounded by butterflies, men carved into impossible muscular perfection, cosmic eyes, neon streets, mystical animals and an apparently inexhaustible supply of golden light.

I recognise the technical achievement. I can also recognise when I have had enough. Earlier generations of image models developed obvious house styles, and some still seem to encourage particular kinds of spectacle. I have experimented with different systems over time, but I repeatedly find myself returning to Flux, with ChatGPT and its image tools increasingly occupying another part of my working process. That preference is partly practical, but it is also aesthetic. Different tools encourage different habits, and I am interested in tools that allow me to move away from their most recognisable defaults rather than towards them. The moment I can immediately identify the generator before I have really looked at the image, something has gone wrong for me. I want to see the work before I see the machine.

3. What is art?

This inevitably leads to a question so large that it ought to come with its own furniture: What is art? I am not sure I can answer it completely, and perhaps anyone who claims to have done so should be treated with mild suspicion. But I can say something about what I need from it.

Art involves intention, but intention alone is not enough. It involves choice, but choosing something does not automatically make the choice interesting. It may involve skill, although skill can appear in strange places and does not always involve making something directly by hand. For me, art begins to emerge when decisions accumulate around an idea.

Why this image? Why this material, or its simulation? Why these colours? Why this sequence? Why eighteen images rather than eighty? Why include one and reject another? What relationship does the work establish between itself and what came before it? What happens when the image is placed beside another image, given a title, written about, returned to, questioned and finally allowed to remain?

These decisions are not glamorous. Most are acts of refusal. That may be especially important in generative art because the machine is exceptionally good at offering more: more detail, more versions, more drama, more polish, another composition, another colour, another body, another object, another hundred possibilities before breakfast.

The artist's role can therefore become strangely inverted. Traditional making is often associated with the difficulty of producing something. Generative making introduces another difficulty: knowing when not to produce. The button is easy. The judgement is the work.

4. Restraint as a position

My own work has increasingly developed through restraint. That does not mean everything has to be pale, minimal or tasteful in some polite decorative sense. I enjoy colour. I enjoy ornament, bodies, pattern, artificial materials, absurdity and visual pleasure. Some of my projects are deliberately excessive. But excess needs a reason.

What I distrust is the automatic spectacular: detail because detail is available, glow because the model understands glow, complexity because complexity looks expensive. The image performs its own technical fluency and expects that fluency to be enough. I want something slower.

I work in projects because a project forces individual images to answer to something larger than themselves. A single generated picture can be seductive very easily. Eighteen related pictures are harder. Repetition exposes weakness. A visual trick that looks wonderful once can become unbearable by image six. A colour that initially seems exciting begins to shout. A composition reveals itself as a habit.

Series demand discipline, but they also create the conditions in which quieter images can survive. Not every picture has to behave like a thumbnail fighting for attention. One image can be sparse because another carries density. A project can breathe.

This matters enormously to me. I want images that can be returned to rather than merely consumed. I want them to feel considered enough to withstand the moment after the initial surprise has gone. The first question therefore becomes less often Can I make this? The tools have made that question almost trivial. The better question is Why should I?

5. Staying off the feed

This may also explain why I have little interest in promoting my work through social media. That is a conscious choice. I understand why artists use those platforms. They offer visibility, community and an audience, and there is nothing inherently wrong with wanting people to see what you make.

But I have no desire to spend half my working life converting the work into content in order to persuade an algorithm to show it to somebody for three seconds. The rhythm feels wrong for me: make, post, check, like, share, repeat.

The image quickly becomes a unit of attention rather than part of a body of work. Success becomes measurable through reaction, which quietly changes the questions an artist asks. What will stop the scroll? What will perform well? What can I post today? What does the platform reward?

I would rather maintain a website. It is unfashionably slow, which may be why I like it. My work sits there in projects. It does not disappear beneath somebody else's breakfast, political argument or sponsored trainers. The viewer can arrive deliberately, look around, read something, leave, return six months later and find the work still where it was. The website gives the images a room.

Perhaps that is what I am trying to protect: not exclusivity, not seriousness for its own sake, simply enough space around an image for attention to become possible.

6. Choosing what survives

I am not opposed to generative abundance. I depend upon it. My work would not exist without systems capable of producing enormous fields of visual possibility. I enjoy their unpredictability, their speed and their ability to take an idea somewhere I had not anticipated. But abundance is raw material, not achievement.

The fact that a machine can generate a beautiful image does not oblige me to keep it. The fact that an image receives attention does not make it meaningful. Technical fluency, novelty and popularity are all real qualities, but none of them settles the question of art.

Perhaps my increasing discomfort with the visual flood has clarified my own position. I do not need to make more images simply because more images are possible. I need to become better at deciding what survives.

That means cultivating taste without pretending taste is universal. It means allowing myself to dislike things without needing to declare them worthless. It means recognising that another artist may want spectacle where I want restraint, speed where I want deliberation, a feed where I want a room. And it means applying the same scepticism to my own work that I apply to everything else.

I make synthetic images in a world already drowning in images. That is not a reason to stop. It is a reason to choose more carefully.

1. A picture has to come from somewhere

We are accustomed to asking where an image came from. A painting points towards a surface and the marks made upon it. However strange or illusionistic the result, somewhere beneath the picture there is material: pigment, canvas, board, paper, brush, hand. A photograph points elsewhere. It begins with something placed before light: a face, a room, a landscape, an object, an event. The image may transform what it records, but there is normally a moment behind it when something occupied the world.

These are very different ways of making pictures, yet both give the image a kind of ancestry. We can imagine a before. Before the painting, there was an empty surface. Before the photograph, there was a scene.

The synthetic image unsettles this arrangement because its before is more difficult to locate. The object may never have been built. The body may never have existed. There may have been no room, no weather, no table, no flower, no moment waiting to be preserved. And yet there is an image, already carrying the signs of things that seem to have happened. That is the strangeness I keep returning to.

2. A different kind of before

When I look at a convincing synthetic portrait, part of me instinctively searches backwards. Who is he? Where was the photograph taken? What kind of light was used? What happened just outside the frame?

These questions arise because the image speaks in a language we already recognise. A particular softness of light suggests a studio. A slight turn of the head implies direction. Skin, fabric and depth seem to occupy the same optical space. Everything encourages the mind to construct an event behind the picture.

But there may be no event to recover. There is no original photograph hidden beneath the image, no sitter who left the studio afterwards, no room waiting somewhere outside the frame. What looks like the residue of an encounter may instead be the beginning of the thing.

This does not mean the image has no history. Quite the opposite. Behind it lies a more diffuse ancestry: photographs, paintings, drawings, advertisements, films, catalogues, diagrams, illustrations, museum displays and countless other ways in which bodies, objects and spaces have already been pictured. The model does not remember these things as a person remembers them, but it has learned from their accumulated appearances.

The synthetic image therefore has a different kind of before: not one subject, one event or one surface, but a field of previous appearances. This reverses a familiar order. Traditionally, the world comes first and the image follows. An artist observes, remembers, constructs or records something, and the picture becomes the consequence. With generative image-making, the image can come first.

A fictional object acquires appearance before it acquires existence — if it ever does. A face appears before there is a person to whom the face could belong. A room can possess afternoon light without ever having passed through an afternoon. The image is not necessarily representing a prior thing. Sometimes it is proposing one.

3. Neither painting nor photograph

This is why I have never found it satisfactory to describe these images simply as a branch of photography, painting or illustration. They can resemble all three, and many other forms besides. A synthetic image may borrow photography's depth of field, painting's surface, drawing's line, cinema's lighting, sculpture's sense of volume or the cool authority of a museum catalogue. It can appear to contain brushwork without paint, graphite without graphite, lens effects without a lens, ceramic without clay.

It moves easily between visual languages that older media had kept materially separate. But resemblance is not identity. A picture that looks painted has not therefore become a painting. An image that behaves photographically has not necessarily been photographed. The appearance of a medium and the physical process of that medium have begun to separate.

This creates an odd category problem. We recognise what we are seeing because its visual grammar is familiar, but the familiar names no longer quite fit. The image may look old while being produced under entirely new conditions. It can wear the clothes of several media without belonging completely to any of them.

That instability interests me more than the argument over whether the result should be allowed into one existing category or another. Perhaps the more useful question is not What older thing is this really? but What becomes possible once an image no longer has to choose?

4. The third image

This is where I think of the third image. The phrase is not intended as a literal chronology. There have obviously been far more than two kinds of image before generative AI appeared. Drawing, printmaking, cinema, collage, digital graphics and countless other forms make any simple history impossible.

The "third" is conceptual. One image is materially made. Another is optically encountered. The third is synthesised into appearance. Painting organises marks. Photography records an encounter with light. The synthetic image establishes conditions from which an appearance can emerge.

What fascinates me is that this third image can carry memories of older forms while remaining materially committed to none of them. It can appear handmade without a hand having made its marks. It can appear witnessed without anyone having witnessed its subject. It can borrow weight from sculpture, surface from painting, authority from photography and atmosphere from cinema, all within the same image. This is not because it stands outside visual history. It is because it is crowded by it.

The third image arrives carrying inherited conventions: ways of lighting a face, displaying an object, composing a landscape, ageing a surface, staging a room. Its novelty does not lie in having escaped everything that came before. It lies in the strange new conditions under which those older visual languages can now be recombined.

That does not make it superior to painting or photography, and it certainly does not make those media obsolete. The opposite may be true. The arrival of another kind of image makes their particular qualities easier to see.

Painting possesses material presence because something was physically done to a surface. Photography possesses the contingency of an encounter because something stood somewhere in time. The synthetic image gives up both certainties, but gains another freedom: the ability to construct appearances without waiting for either material fabrication or photographic occurrence. Its limitation and its possibility are the same: there is no single original behind it.

5. Possibility as material

This changes the artist's relationship with the image. When the world does not have to provide the subject, possibility becomes unusually abundant. The image can move towards things that could exist, things that almost exist, things that cannot exist, or ordinary things that simply never did.

But infinite possibility is not the same as artistic possibility. A system capable of generating almost anything can just as easily produce endless variations of nothing in particular. The absence of material or photographic limits does not remove the need for limits. It makes them more important.

So the interesting part begins with restriction. A project establishes a territory. Certain colours belong and others do not. A material has to behave in a particular way. One image is accepted and another refused. A sequence develops its own internal pressure. What initially exists as possibility gradually becomes specific. This feels different from facing a blank canvas, but it is also different from pointing a camera towards the world.

The blank canvas begins with absence and waits for marks. The camera begins with presence and selects from it. The synthetic image begins with potential. The work is to give that potential form.

6. An image made from images

There is another peculiarity to this condition. The synthetic image is not only an image of a possible thing. It is also, in some sense, an image made through the accumulated memory of other images.

That does not mean it simply pastes fragments together or retrieves a hidden picture from an archive. The process is far stranger and less literal than that. But the visual languages it can summon have been learned from a culture already saturated with representation.

This means the third image is both new and inherited. Its portrait may contain the distant habits of fashion photography, studio lighting, classical portraiture, advertising and cinema without belonging directly to any one of them. Its still life may suggest Dutch painting, product photography and catalogue display simultaneously. A fictional sculpture may appear already museum-ready because the image has inherited not only the appearance of sculpture, but the visual conventions through which sculpture is normally photographed and presented.

The image does not merely depict an object. It may also carry the history of how objects like that have been shown to us. This is where the synthetic image becomes especially strange. It can simulate not only a thing, but the cultural memory surrounding the thing. A vase may arrive with the ghost of the museum already inside it. A portrait may arrive already knowing how portraits are supposed to behave. A room may seem to remember photographs of rooms that never existed within it. The third image is therefore not empty space waiting to be filled. It is already inhabited.

7. Before the definition

Eventually, a medium requires language around itself. Names matter because they create distinctions. They allow a practice to become discussable rather than remaining permanently attached to the novelty of its tools. Photography had to become more than "pictures made by a machine". Cinema became more than moving photography. Digital art eventually became too broad a territory to describe every practice taking place inside a computer.

Generative image-making is passing through a similar period of uncertainty. "AI art" may describe the technology involved, but it says very little about the kind of image being made, the artist's intentions or the discipline surrounding the result.

Elsewhere, I use the word synthography for the practice I have developed around these images. There are practical and linguistic reasons for that name, and they deserve their own explanation. But before definition comes recognition.

Something has changed in the relationship between picture and world. An image can now possess the appearance of a history it never lived. It can look observed without having been witnessed, handmade without having been touched, old without having existed yesterday. At the same time, it does not arrive empty-handed. Photography, painting, drawing, cinema, design, sculpture and the archive remain present within it as inherited ways of seeing.

The third image is therefore neither rootless nor easily placed. It has no single origin, but many ancestors. It can inherit from photography without being a photograph and from painting without being painted. It can carry the memory of older media while remaining materially separate from them.

That does not make it an empty imitation of what came before. It gives us another kind of image to think with. The third image does not replace the pictures that preceded it. It appears among them — strange, familiar, crowded with memory — and quietly asks us to make room.

1. Matter without matter

A strange thing happens when I make a synthetic image of an object. Glass appears, but no silica has been melted. Timber carries grain although no tree was felled. Ceramic seems glazed without clay, kiln or heat. Metal acquires scratches, oxidation and weight without ever becoming solid enough to touch. A flower bends towards imaginary light although it has never grown. The image presents the consequences of matter without possessing matter itself.

This contradiction sits close to the centre of synthography. The things I make often depend upon physical credibility: the thickness of a glass rim, the dullness of weathered timber, the softness of cloth, the unevenness of fired clay, the way metal catches light along an edge. Yet none of these materials exists within the process. There is no studio table covered in tools and scraps. No sawdust. No fingerprints in wet clay. No smell of paint, hot metal or timber. There is only an image in which those things appear to have happened. I think of this as a kind of digital alchemy.

Alchemy traditionally imagined transformation: one substance becoming another, ordinary matter acquiring extraordinary properties. Synthography performs a different sort of transmutation. It does not transform lead into gold. It transforms language, computation and probability into the appearance of material experience. Something immaterial begins to behave as though it had been made.

2. Language as material

It is tempting to say that a synthographer makes images from words. That is true, but only partly. Words initiate the process. I can describe a vessel as translucent, hand-blown, imperfectly circular, scratched with use and photographed in soft window light. I can ask for salvaged timber, tarnished brass, chalk on brown paper or a body emerging from charcoal darkness. Language establishes conditions: material, scale, atmosphere, age, colour, construction and light. But the words themselves do not contain the finished image.

Between instruction and appearance lies the generative system: a vast field of learned visual relationships capable of translating linguistic suggestion into form. The phrase weathered oak does not contain a piece of timber. It activates possibilities associated with grain, colour, damage, joinery, age and surface. Frosted glass does not describe every refraction or highlight. It points towards a visual behaviour.

The image arrives somewhere between what I have asked for, what the model associates with those words and what I am willing to accept. This is why prompting alone has never felt sufficient as a description of the work. Language opens the door, but judgement decides what is allowed through it.

A generated surface may be beautiful and still feel wrong. A wooden cabinet may possess exquisite grain but impossible joints. A metal flower may look intricate but have no convincing sense of how its petals were formed. A ceramic vessel may carry cracks that appear decorative rather than structural. The machine can imitate the appearance of material astonishingly well. It cannot decide whether the imitation makes sense. That remains part of the artist's work.

3. The imaginary workshop

Much of my practice therefore takes place in an imaginary workshop. Nothing is physically fabricated there, yet fabrication matters constantly.

If I ask for a vessel that appears handmade, I begin thinking about thickness, gravity, irregularity and the marks left by process. If I make a fictional cabinet from reclaimed timber, I have to consider how one board meets another, where a hinge belongs, how a handle might function and whether the object looks capable of standing under its own weight. The object does not have to exist. But it has to understand, visually, what existence would require.

This is one of the pleasures of synthetic image-making. Processes that would normally require entirely different studios can occupy the same digital space. Glassblowing, ceramics, metalwork, photography, drawing, textile design, sculpture and carpentry become neighbouring languages. I can move between them without owning a furnace, kiln, press, workshop or photographic studio. Yet their histories of making still matter.

A synthetic ceramic surface convinces because it carries signs we associate with ceramics. A charcoal drawing needs pressure, dust, erasure and the uneven drag of material across paper. A piece of oxidised metal needs more than a coloured texture laid over a shape; the corrosion has to appear to belong to the surface and to time. The processes are fictional, but their visual logic cannot be arbitrary.

Perhaps this is why I return so often to materials. They give the artificial image resistance. They force it to account for itself.

4. The discipline of material behaviour

Detail alone is not enough. Generative systems are extremely good at detail. They can produce pores, scratches, fibres, cracks, grain, patina and reflection in almost indecent quantities. But excessive detail can make an image less believable rather than more. Matter is not convincing because everything has been described. It is convincing because things behave consistently.

Glass has to transmit and distort light rather than merely sparkle. Cloth has to fold according to weight and tension. Timber needs grain that follows form. Skin requires transitions rather than a catalogue of pores. Metal should reflect its surroundings differently from ceramic or stone. These differences are small, but they are where material credibility lives.

For me, this is where selection becomes more important than generation. The first image is rarely the answer. Something may be too glossy, too perfect, too complicated or simply too eager to demonstrate how impressively artificial intelligence can render a surface. I often find myself removing spectacle rather than adding it. The aim is not maximum realism. It is enough coherence for the material fiction to hold.

When that happens, an odd perceptual threshold is crossed. I know that the ceramic vessel has never been fired, yet I can almost imagine picking it up. I know that the wood has never grown, yet its grain suggests direction and resistance. I know that the drawn line contains no graphite, yet I read pressure in it. The image has acquired a memory of touch without ever having been touched. That interests me far more than technical perfection.

5. Digital alchemy

So when I use the phrase digital alchemy, I do not mean that artificial intelligence performs magic. If anything, the interesting part begins when the magic trick becomes visible.

A synthographic image can imitate the outward evidence of physical processes while remaining entirely apart from them. It can carry the appearance of labour without reproducing that labour, the appearance of matter without possessing matter, the appearance of history without having passed through time. That distance is not something I want to conceal. It is the territory of the work.

The glass cannot break. The flower cannot wilt. The metal will never tarnish beyond the tarnish already contained in the image. The fictional cabinet has no back unless the image gives it one. These objects exist only from the position in which they have been made visible. And yet they are not nothing.

They have colour, proportion, rhythm, association and presence. They can be examined, compared, arranged into sequences and returned to. They can evoke the memory of objects without becoming objects themselves. What has been transformed, then, is not language into matter. It is language into the experience of matter through an image. That distinction is small, but important.

Synthography does not manufacture the world. It manufactures appearances capable of making us remember how the world feels. That is its peculiar alchemy: no gold, no glass, no wood, no flesh, no paper — and nevertheless, before the eye, something seems to have taken form.

1. The face is not another surface

Generative AI can imitate almost anything. Glass can acquire thickness and refraction. Timber develops grain. Metal tarnishes. Fabric folds under imaginary weight. A ceramic vessel can appear to have passed through a kiln that never existed. The human face is different.

A face is not simply a collection of surfaces arranged convincingly under light. We read intention into it. We look for character, mood, attention and recognition. A slight movement around the mouth can suggest reserve or amusement. The direction of the eyes can make us feel noticed or ignored. We respond to faces socially before we begin analysing how they have been constructed.

That made synthetic portraiture difficult for me long after artificial intelligence had become capable of making impressive images. Earlier systems could generate attractive men with apparently realistic skin, hair and bodies, but something often failed beneath the surface. Hands were unreliable, anatomy drifted, expressions seemed vacant and faces tended towards a polished generic beauty. The image might persuade for a moment, but sustained attention revealed the construction.

The men were convincing enough to look human, but not specific enough to feel like individuals. They were perfect strangers.

2. A disposable beauty

There was another problem. Generate a handsome man, then generate him again, and he was usually gone. The next image might contain someone similar — the same general age, haircut, physique or mood — but not the same person. A face existed for the duration of a single picture and then dissolved back into possibility. This gave early synthetic portraiture a peculiar disposability.

The subject could appear intensely present while you looked at him, yet he possessed almost no continuity beyond the frame. There was no before or after, but there was also no reliable next time. The image produced a person-shaped event rather than a persistent identity.

That instability affected the way I worked with male figures. Projects such as Gymnos and Kalos were openly concerned with beauty, desire, the classical body and the erotic charge of looking, but I became increasingly aware of how easily generative systems could flatten those concerns into an endless supply of attractive men: youthful, muscular, symmetrical, available. Another one could always be generated.

The ethical question was not whether these fictional men had consented to appear. There was nobody there to give or withhold consent. The more useful question concerned my own decisions. What kinds of bodies was I repeatedly asking the machine to produce? What inherited ideas of masculinity and desirability was I reinforcing? Was the figure functioning as part of an artistic argument, or had he become simply another beautiful surface? The ease of production made those questions harder to ignore.

3. When the image looks back

Then the technology improved. The change was obvious in anatomy and detail, but those were not the things that interested me most. Better hands are useful. More credible skin is useful. Bodies that obey gravity are useful. What changed the experience was presence.

In later work, particularly with Flux.2 Pro, the synthetic men began to occupy their images differently. Expressions became less fixed. Posture seemed to carry intention. Skin could contain warmth and irregularity without turning into a demonstration of pores. Light appeared to settle across the body rather than merely coat it. The figures began to feel observed rather than assembled.

My Garçon project developed from that discovery. The portraits were intentionally restrained: simple clothing, quiet backgrounds, controlled light, little narrative distraction. I wanted to know whether a fictional sitter could survive the discipline of ordinary portraiture without needing fantasy, spectacle or obvious artificiality to hold the image together. Something unexpected happened. The men began to look back.

Of course, they were not actually looking at anything. There was no consciousness behind the eyes and no sitter responding to the camera. But the visual conditions of portraiture were strong enough to produce the sensation of encounter. The artificial figure had not become human. The image had become socially persuasive.

4. The same man again

More recently, another threshold has appeared. The question is no longer simply whether artificial intelligence can generate a convincing person. It is whether it can generate the same person again.

Reference-based image generation increasingly allows facial structure and other identifying characteristics to persist while clothing, pose, framing and setting change. That technical development may sound modest beside dramatic improvements in photorealism, but for portraiture it changes something fundamental. Continuity creates identity.

In MODEL, I deliberately tested this by constructing a project around one invented man. Across eighteen images, almost everything around him could change, but the man himself had to remain recognisable. This turned generation into a strange form of preservation.

The challenge was no longer to make a good-looking portrait. It was to prevent the subject from drifting into somebody else. A slightly altered jaw mattered. The wrong eyes mattered. A change in skull shape mattered.

At some point I realised I had developed a surprisingly firm idea of what this nonexistent man was supposed to look like. An image could be technically excellent and I would reject it for a very simple reason: That isn't him.

5. Becoming familiar

This is where synthetic portraiture becomes psychologically interesting to me. Human beings are extremely good at becoming familiar with faces. We learn to recognise people across changes in lighting, expression, hairstyle, age and viewpoint. Once someone is familiar, we no longer require the exact same photograph in order to know who we are looking at.

A recurring synthetic figure begins to exploit the same perceptual habit. The first image introduces a face. The second confirms it. By the fifth or tenth appearance, recognition begins to operate before analysis. You know him.

Not personally, of course. There is no personality hidden behind the image and no private life taking place when you are not looking. But visually, a relationship has formed. Certain features have become expected. Difference becomes noticeable because sameness has been established.

The strangest moment comes when the figure seems to require a name. Names belong to individuals. We do not normally name every generic face we encounter in an advertisement or illustration. A name becomes tempting when recurrence has made the figure specific enough to distinguish from others.

That does not create personhood, but it does reveal how readily personhood begins to gather around consistency. The machine provides the recurring face. The viewer begins supplying the social meaning.

6. The stranger remains

This creates a peculiar kind of portrait. The synthetic sitter can become visually durable without becoming alive. He can acquire continuity without biography, familiarity without memory and presence without consciousness. He may appear repeatedly in different clothes and rooms. He can seem older, tired, amused or distracted. He may appear to possess moods, preferences and habits. But all of those qualities belong to the images. There is no hidden life connecting one frame to another. That is why I still think of him as a stranger.

The achievement of contemporary generative portraiture is not that it has manufactured a person. I do not believe there is a little artificial soul waiting behind the pixels for someone to notice it. Something subtler has happened.

The technology has become capable of maintaining enough visual coherence for us to begin doing what we naturally do with human faces: recognise, remember, infer, project and become familiar.

This changes my relationship with the figures I make. They are no longer simply interchangeable embodiments of an idea. A recurring face can begin to impose its own constraints upon a project. I find myself protecting its identity, noticing deviations, asking whether an expression feels right for him rather than merely whether it is well rendered. That "him" is entirely constructed, and yet the pronoun arrives almost automatically.

Perhaps that is where the perfect stranger now stands. He does not exist, but he can recur. He has no past, but he can develop a visual history. He cannot recognise me, but I can recognise him. And once that happens, the image no longer feels quite as empty as it did before.

1. Looking at oneself

The desire to look at ourselves is much older than the smartphone. Long before the front-facing camera, people used mirrors, paintings and drawings to study and represent their own appearance. Self-portraiture has always contained something slightly strange: the person making the image becomes both observer and subject, looking outward and inward at the same time.

Egon Schiele's self-portraits make this tension particularly visible. Between 1910 and 1918 he produced a large series of psychologically intense images of himself, often using his own body as a site of distortion, exposure and performance. Schiele was not making selfies in any literal historical sense, but his work reminds us that bodily self-scrutiny, erotic self-presentation and the performance of identity did not begin with social media.

Photography itself was barely born before somebody turned it upon himself. In 1839, only months after Daguerre's photographic process had been announced publicly in Paris, Robert Cornelius made a daguerreotype self-portrait in Philadelphia. The exposure took several minutes.

There is something wonderfully contemporary about that gesture despite the cumbersome technology. A new image-making system appears, and almost immediately someone asks: what do I look like in it?

The technology changes, but the desire does not. What interests me is what happens when technology progressively removes the distance between having a body, looking at that body, making an image of it and showing that image to somebody else.

2. The body becomes an image

A mirror gives us an immediate image of ourselves, but it disappears when we walk away. Photography changed that by making the reflected or observed body durable. An expression, pose or gesture could be fixed, inspected later and shown to somebody who had not been present when the image was made. The body acquired a double.

There was the physical body, changing continuously in time, and there was the photographed body: still, framed and separated from the person who produced it. Once that double existed, it could be studied in a way that a passing reflection could not. Do I look attractive? Strong? Awkward? Is that really what other people see?

For most of photography's history, however, this process contained considerable friction. Early portrait photography required long exposures. Later film made the process much easier, but exposures were still finite, the result remained invisible until development, and processing often happened somewhere else. The photograph returned after the moment that produced it had passed. The mirror was instant but temporary. Photography was durable but delayed. The history that follows is partly the history of that delay disappearing.

3. Closing the gap

A major shift arrived with instant photography. Polaroid's first commercial instant camera, the Model 95, went on sale in 1948, allowing the photograph to be developed as part of the photographic process rather than being sent away to a conventional processing laboratory.

For self-portraiture, the important change was not simply convenience. It was feedback. The person taking the photograph could see the result while the circumstances that produced it were still fresh. The image could begin influencing the next image. Move the shoulder. Turn towards the light. Stand differently. Take another.

When photographer and sitter are the same person, this becomes a loop: the body performs, the image returns, the performance is adjusted. Film photography had created a body double. Instant photography allowed that double to begin talking back.

Electronic imaging tightened the loop further. In 1975, Kodak engineer Steve Sasson built an early digital camera prototype. The technology would take years to become a normal part of consumer life, but its long-term effect was profound. Photography was gradually separated from film, chemical development and the individual physical print.

Once digital cameras acquired screens, the delay between exposure and inspection almost vanished. A photograph could be reviewed immediately and deleted just as quickly. Taking another image cost effectively nothing. Ten photographs became easier than one. A hundred became possible. The image became less precious individually, but vastly more useful as a tool for examining and adjusting oneself.

4. The digital mirror

The next transformation was not simply another kind of camera. It was the convergence of several technologies into one object. When Apple introduced the first iPhone in 2007, it combined a camera and photo library with email, web access and mobile communication in the same handheld device. The original iPhone was not yet the perfect selfie machine — it lacked the front-facing camera that later became so familiar — but the direction of travel was clear. Camera, screen, communications device and network access were beginning to occupy the same pocket.

In 2010, the iPhone 4 added a front-facing camera and FaceTime. The screen could now function much more directly like a mirror while the device simultaneously acted as the camera. That combination matters.

The modern smartphone is camera, mirror, archive, editing suite, communication system and publishing platform at the same time. An image can be made, inspected, deleted, remade, altered and distributed without changing devices or locations.

The body can therefore exist within a continuous visual feedback loop. We see ourselves, photograph ourselves, judge what we see, modify the pose and try again. The image is no longer simply a record of how we looked at one moment. It becomes part of how we decide to look.

The gym mirror makes this especially obvious. The photograph may begin as documentation: progress, muscle gain, weight loss, training, posture. But the same image can also become performance. Lighting is noticed. The abdomen tightens. The arm moves slightly away from the torso. The phone hides part of the face. What appears casual develops its own visual grammar. The body is being recorded, but it is also being directed.

5. The body enters the network

The smartphone alters self-photography again because the image no longer has to stop at the device. A photograph can pass almost immediately from private inspection into messaging, dating platforms, fitness communities and social media. The act of looking at oneself becomes connected to the possibility of being looked at by others. That changes the image before it has even been taken.

A gym photograph can function as evidence of discipline, sexual display, personal record, flirtation, advertisement or simple vanity. Often it performs several of these roles simultaneously. Visual codes associated at different times with bodybuilding, fashion, pornography, gay male imagery and commercial fitness culture now circulate freely through mainstream male self-presentation. The boundaries are porous.

A straight man may produce a carefully lit photograph of his torso for a fitness audience. A gay man may employ almost identical visual language on a dating profile. An athlete or fitness influencer can turn the same gesture into branding. The image itself does not necessarily tell us what form of desire, aspiration or commerce produced it.

What matters is that men have become increasingly accustomed to constructing themselves as images. The camera no longer simply records masculinity. It participates in its performance.

The imagined audience can now be present before the photograph exists. The question shifts from How do I look? towards How will I look when somebody else sees this?

6. Body Double

Technology did not invent narcissism, vanity, self-scrutiny or the desire to be desired. It gave those impulses increasingly efficient machinery.

The progression is remarkably clear. The mirror allowed the body to see itself but could not preserve the reflection. Photography preserved it but introduced delay. Instant photography shortened that delay. Digital imaging made repetition cheap and immediate. The smartphone brought camera and mirror together with editing, storage and communication. Networked culture supplied an audience. At every stage, the distance between body and image became smaller.

Yet something curious happens as that distance contracts: the image begins to influence the body that produced it. People train for photographs, pose for photographs, dress for photographs and alter themselves because of photographs. A body may be built partly in response to the image it hopes eventually to become. The double starts shaping the original.

This is where the modern self-portrait becomes more than documentation. It enters a loop between appearance and aspiration, between the body as lived and the body as seen. The photograph is no longer simply evidence of the body. It can become one of the instruments through which the body is imagined and constructed.

I find that loop fascinating because it is neither entirely superficial nor entirely profound. Sometimes a selfie is simply a selfie. Sometimes looking at oneself is vanity. Sometimes it is curiosity, sexuality, insecurity, pride, evidence, play or the simple pleasure of recognising your own body.

Human beings have always wanted to see themselves. Robert Cornelius was already standing in front of his camera in 1839. Schiele was examining and performing his own body on paper decades later. Polaroid compressed the wait. Digital photography almost eliminated it. The smartphone placed the camera, mirror and audience in the same hand.

What is new is not the desire. It is the speed, ease and frequency with which that desire can become an image — and the speed with which the image can return to influence the person who made it. The mirror once returned our gaze and disappeared. The body double remains.

1. The old contract

For almost two centuries, photography has carried a particular relationship with the world. A photograph may be staged, cropped, retouched, manipulated or deliberately misleading, but traditionally it begins with an encounter between light and something that was there. A person stands before a lens. A building occupies a place. Light reflects from a surface and reaches film, photographic paper or a digital sensor. Something happens in front of the apparatus, however carefully that happening has been arranged.

This does not make photographs automatically truthful. Photography has never been innocent. Portraitists direct their sitters, advertisers manufacture desire, governments select images for political purposes, and photographers decide where to stand, when to press the shutter and what to exclude from the frame. Yet beneath all these choices lies a physical connection between photograph and subject. One influential way of describing this is to call the photograph indexical: the image bears a trace of something that was present when it was made. The relationship is not proof of truth, but it is a relationship to an occurrence.

Even the camera itself is not essential to this principle. Photograms can be made by placing objects directly onto photosensitive material and exposing them to light, producing photographs without lenses or cameras. What matters is not the familiar black box but exposure: light from something present acts upon a sensitive surface.

That physical encounter has helped shape photography's cultural authority. We know photographs can lie, but we also know that, historically, something normally had to be there in order for the photograph to begin.

2. From chemistry to computation

Photography has changed technologically many times without entirely abandoning this relationship. Glass plates gave way to film. Black-and-white expanded into colour. Chemical processing was joined and eventually displaced in everyday practice by digital sensors, software and smartphones. The material structure of the photograph changed enormously, but digital photography still began with captured light. The sensor replaced the light-sensitive film emulsion, and software increasingly interpreted what the sensor received, yet the image remained connected to a moment in front of the device.

Digital photography was sometimes treated as though it had broken photography's relationship to reality. Pixels seemed more unstable than negatives. Digital files could be copied perfectly and altered invisibly. Photoshop made manipulation easier and more familiar. The modern smartphone complicates things further by processing photographs computationally before we have even seen them. Yet processing is not the same as generation.

A heavily processed digital photograph may have been sharpened, merged from several exposures, colour-corrected, blurred, brightened or selectively reconstructed, but photographic information still entered through an optical encounter with the world. The resulting image may be highly mediated, but its starting point remains an exposure. Generative AI introduces a different condition. It can begin without an exposure at all.

3. When the photograph disappears

A synthetic image can now display almost every familiar characteristic of photography without having been photographed. It can contain apparent depth of field, lens behaviour, studio lighting, skin texture, motion blur, reflections, shadows and the tiny imperfections through which photographs acquire credibility. A fictional man can appear to have stood beneath a softbox. An imaginary room can seem photographed in afternoon light. An object that has never been manufactured can sit convincingly on a tabletop and cast a shadow onto a surface that is equally fictional. Nothing needed to stand before a lens.

This is the break that interests me most. Generative image-making does not simply give us another method for manipulating photographs. It allows photographic appearance to become independent of photography. Photorealism can now be produced without the direct relationship to an event or subject that traditionally belonged to the photograph.

The difference sounds technical, but culturally it is enormous. For much of photography's history, if an image possessed the visual grammar of a photograph, there was usually some photographic event behind it. Perhaps the event had been staged. Perhaps the negative had been retouched. Perhaps the photograph was propaganda. But somewhere in its history there was normally a subject, a light-sensitive surface and a moment of exposure.

Now the visual consequences of that event can be synthesised without the event itself. We have entered a period in which an image can look photographed without being a photograph.

4. Portraiture without a sitter

I became particularly conscious of this through portraiture. For a long time I avoided making realistic human figures with generative AI. Earlier models could produce striking faces, but the illusion often collapsed around them: awkward anatomy, unreliable hands, waxy skin, vacant expressions and a general sense that the person had been assembled rather than observed.

As the technology improved, that changed. Working with Flux.2 Pro, I found myself able to make images possessing the discipline and restraint I associated with studio portraiture. A man, a lens, a moment of stillness — except there was no lens in the photographic sense, and there was no man. I was not recording a sitter or waiting for light to fall across an existing face. I was directing the appearance of both sitter and light into being.

That shift affected the way I thought about the resulting images. Flux.1 often seemed to give me images of men. With Flux.2, I began to feel that I was seeing men in images. The distinction is subjective, but important to my practice.

The figures remain synthetic. They have no childhood, no physical body, no morning before the portrait session and no life afterwards. Yet improvements in anatomy, expression, skin, posture and light allow them to occupy the image with something approaching presence. In my Garçon project I began treating these fictional sitters according to the conventions of serious portraiture: restrained clothing, controlled light, uncluttered backgrounds, attention to posture and an emphasis on looking rather than spectacle.

The result raises an uncomfortable but productive question. If portraiture has historically involved an encounter between photographer and sitter, what kind of portrait is produced when one half of that encounter has disappeared?

I do not think the answer is that synthetic portraiture has simply become photography. It has become capable of borrowing photography's language with extraordinary fluency.

5. What photography keeps

This distinction makes me more interested in photography, not less. Photography still possesses qualities that generative imagery cannot reproduce because they depend upon an event actually taking place. A photographer can encounter something unexpected. A sitter can resist direction. An expression can change between one exposure and the next. Weather can intervene. Somebody can walk unexpectedly into the frame. The light can disappear. The photographer can misunderstand the situation and discover later that the unsuccessful exposure contains something more interesting than the planned one. Photography deals with contingency. It also deals with another person's presence.

Portrait photography is never only the production of a face. Even in highly controlled commercial work, there is a negotiation between photographer, sitter, camera, space and time. The sitter brings a body that becomes tired, uncomfortable, confident, embarrassed or bored. They possess an identity beyond the frame and can participate in how that identity is represented. The finished portrait records not simply an appearance but the residue of an encounter.

This is why photography remains particularly powerful as witness. Its relationship to an event continues to matter in documentary practice, journalism, archives, family photographs, scientific records and countless everyday uses where the fact that something happened is part of what gives the image meaning.

A synthographic portrait can simulate the nervous glance. It cannot have witnessed it.

6. A camera for ideas

Where synthography changes the field is not by removing these qualities from photography, but by revealing that they are no longer necessary for every kind of photographic-looking image.

I can now construct light without lamps, models without bodies and locations without geography. I can work with the visual vocabulary of commercial photography without hiring a studio, building a set or organising a shoot. In practical terms this is liberating. An individual artist can explore kinds of imagery that once required substantial budgets, equipment and coordination. But convenience is only the surface of the change.

More fundamentally, generative AI separates two things that photography had kept closely connected: the photographic look and the photographic event. Once separated, each becomes easier to see.

Photography's distinctive quality is no longer simply its ability to produce convincing images. Synthetic systems can do that too. Its distinctiveness increasingly lies in its relationship to what happened: somebody was there, light fell, time passed, the shutter opened, the world interfered.

Synthography occupies the other side of that division. Its strength lies precisely in not having to wait for an event. It can begin with an idea and build towards an image. Where photography is constrained and enriched by what stands before it, synthography is constrained and enriched by what can be imagined, directed and made coherent.

Photography records possibility after it becomes real. Synthography can picture possibility before reality is involved at all.

7. What changes now

This transformation inevitably extends beyond individual artists. Photography is also a profession and an industry. Advertising, fashion, editorial work, product photography and commercial portraiture have traditionally depended upon networks of photographers, models, assistants, stylists, retouchers, producers, studios and locations. When convincing imagery can be generated without some or all of that infrastructure, economic pressure follows. The anxiety surrounding these technologies is therefore not imaginary, nor should it be dismissed as simple resistance to change.

New image-making systems alter who gets commissioned, what kinds of labour are required and which skills retain economic value. Generative AI is not merely another editing tool; it changes parts of the production process itself.

At the same time, I do not believe photography simply ends because another medium can imitate its appearance. Painting did not disappear when photography arrived. Photography did not disappear when digital imaging arrived. Technologies change the conditions under which older media operate, sometimes painfully, and force distinctions that previously seemed unnecessary.

Generative AI may do something similar. Photographers may increasingly find that the value of photography lies not in producing an image that looks photographic, but in what only photography can bring to that image: access, encounter, testimony, collaboration, unpredictability, lived time and the knowledge that a particular person or thing stood somewhere in the world.

Synthetic image-makers face the opposite responsibility. If the world is no longer required to provide the subject, then intention has to provide a reason for making one. The freedom is enormous, but so is the obligation to use it well.

8. The camera never lies

The old phrase the camera never lies was never really true. Cameras have participated in lies almost from the beginning because people decide what cameras are shown, what they exclude and how their images are presented. A photograph can deceive while remaining entirely photographic. But there is another sense in which the phrase now seems newly interesting.

The camera cannot photograph something that was never there. A generative system can show it anyway. That does not make one medium truthful and the other false. It means their relationships to reality are fundamentally different.

My tools now allow me to make images that possess the surface authority of photography while beginning somewhere else entirely: in language, intention, probability and selection. They do not replace the camera. They expose more clearly what the camera was doing all along.

Photography is an encounter with something that has entered the world. Synthography is an encounter with something that may never need to. AI has not ended photography. It has made photography more specific. And perhaps that is how it reshapes it most profoundly.