What has had the greatest impact on the world over the past three years? Setting aside a handful of autocrats, it is without doubt artificial intelligence. Those autocrats have an expiry date, but AI is here to stay and is developing exponentially. It has an unimaginable influence on virtually every conceivable field.

Photography, web design and the arts are no exception. I therefore try to keep up with developments as best I can. This week I read an essay in Andy Adams' Flak Photo newsletter. The essay was written by photographer Peter Hoffman, who put forward the following intriguing argument: that the punctum is impossible in an AI-generated image. I have touched on this before, but what exactly is the punctum?

In 1980, French theorist Roland Barthes published a slim volume called Camera Lucida. Since then it has become required reading for many photography students. Barthes introduced the concepts of the studium and the punctum, where the studium refers to what is objectively visible in a photograph. The punctum, by contrast, is a small, unintended detail in the image that strikes the viewer in a personal and unexpected way.

Imagine: you see an old photograph of an unknown woman from the 1930s. She stands formally dressed in front of a house. That is what the photograph wants to show and what Barthes calls the studium, the deliberate subject.

But then you notice her shoes. A lace is undone, and the shoes are just slightly too large. Nobody paid attention to that when the photograph was taken. Yet that detail pricks you. It touches you, perhaps because you suddenly see an entire life in it: haste, poverty, vanity, or simply a morning when she hadn't been paying attention.

That is the punctum.

Now the crucial point: why does it work that way? Because you know, and this is almost physical, that those shoes were really there. That that woman really stood on that ground. The light truly fell on her feet and was recorded. The photograph is like a fingerprint of reality. There is a direct, physical connection between that moment and the image you are looking at now.

That is what Barthes means by emanation: the image radiates something of the thing itself, the way a glowing object radiates heat. Not symbolically, but almost literally.

And that is precisely why, Hoffman argues, an AI image cannot do this. If you see a shoe with an undone lace in an AI image, those shoes never existed. There was no morning, no woman, no haste. The machine combined pixels on the basis of patterns. There is nothing to return to. The thread to reality simply does not exist.

The punctum pulls on that thread. If there is no thread, there is nothing to pull.

At first glance, Hoffman's argument is convincing. That the punctum requires an ontological bond between image and reality, that is no rhetorical trick, but a serious philosophical observation. And yet, much as I would like to believe Hoffman's story, something about it pinches, like a proverbial ill-fitting shoe.

Firstly, his reasoning about AI is circular: AI images have no punctum because they are not human, and they are not human because they cannot have a punctum. Secondly, he does not actually demonstrate why an unexpected detail in an AI image cannot affect a viewer just as powerfully. He merely asserts it.

And at that point, as a former researcher, I feel compelled to investigate further with at least a small sample.

First: can AI even recognise a punctum?

The answer appears to be yes.

The punctum in my image Querencia, as indicated by Claude

This week I made this photograph in my studio, titled 'Querencia'. It is a Spanish concept meaning something like "a place where you feel deeply at home, where you draw strength, where you are yourself. Not just a favourite place, but a place that nourishes your soul and to which you always long to return". I submitted it to ChatGPT, Claude, and Gemini. All three identified the following punctum (indicated by the arrow), and Claude elaborated as follows:

"For Barthes, the punctum is what pricks — not what the image intends to show, but what strikes you without your asking for it.

In this image, the punctum for me is not the flower itself — that is too deliberately composed, too studium. It lies in the shadow.

The shadow beneath the bird nest has something unexpected: it does not recede as a decorative element, but seems to double the thing as a negative of itself — like an echo, or a grave. The object floats above it, loose, almost hovering. But the shadow pulls it back to earth.

In this there is something of what Barthes described as ça a été — this has been. The dry, dead flower head (an spent allium, I suspect) nests in something that is also a nest or a ruin. Beauty that has already outlived itself.

The punctum for me, then, is: the combination of floating and falling at once — the object lifts free of the ground, but its shadow will not let it go."


Again, the punctum is by definition subjective and may differ between viewer and maker, but in my view that is a remarkably perceptive and well-articulated reading.

Then a second test.

Can AI itself generate an image with a clear punctum?

Gemini 3/Nano banana currently appears to be the best model for image generation. I therefore asked it to create an image with a distinct punctum.

Image generated by Gemini

Gemini identified three locations in the image as possible punctum:

1. The reading glasses hanging on a cord

For me the punctum lies in the glasses resting low on the woman's chest. While the studium of the photograph concerns old age, solitude and nostalgia, the glasses are a very specific, almost painfully tangible detail. They suggest an interruption: she had been reading or doing needlework, but looked up. It is an object that emphasises her physical presence and her 'being in time'. It is not a 'beautiful' element of the composition, but a human, slightly awkward detail that pulls the photograph out of the purely aesthetic.

2. The slight wear on the armrest

Another punctum could be the worn fabric on the edge of the armchair. It tells a story of years of sitting in exactly the same spot. It is a detail that testifies to inexorable time (Chronos), which Barthes often associated with death and the 'that-which-has-been'.

3. The position of the hands

Look at the way her hands are clasped together — not forcefully, but resting. The faint glint of a wedding ring just visible may serve as the punctum for a viewer: a small detail that suddenly sharpens the entire emotional charge of the image (loss, fidelity, survival).

Well, that wedding ring. That's it, as far as I'm concerned.

With this small sample of my own I cannot of course conclude that we as creative professionals have all been made redundant, but I am unfortunately also not yet fully reassured by Hoffman's essay.

Perhaps the most important remaining question is whether the punctum resides in the image or in the viewer. My sample suggests the latter.


AI’s punctum

Wat heeft de afgelopen drie jaar de grootste impact op de wereld gemaakt? Een stelletje autocraten buiten beschouwing gelaten is dat zonder twijfel kunstmatige intelligentie. Op die autocraten zit een houdbaarheidsdatum, maar AI gaat niet meer weg en ontwikkelt zich exponentieel. Het heeft een onvoorstelbare invloed op navenant elk denkbaar vakgebied.

Zo ook voor de fotografie, webdesign en de kunst. Ik probeer de ontwikkelingen daarom zo goed mogelijk te volgen. Deze week las ik een essay in Andy Adam’s Flak Photo newsletter. Het essay was geschreven door de fotograaf Peter Hoffman, met de volgende interessante stelling. Hoffman betoogde dat het punctum onmogelijk is in een AI-gegeneerd beeld. Ik heb het er al vaker over gehad, maar wat is dat ook alweer, het punctum?

In 1980, publiceerde de Franse theoreticus Roland Barthes een boekje genaamd Camera Lucida. Sindsdien is het verplichte kost voor veel fotografiestudenten. Barthes introduceerde de concepten van het studium en het punctum, waarbij het studium aanduidt wat er objectief te zien is in een foto. Het punctum daarentegen is een klein, onbedoeld detail in het beeld dat de kijker persoonlijk en onverwacht raakt.

Stel je voor: je ziet een oude foto van een onbekende vrouw uit de jaren dertig. Ze staat formeel gekleed voor een huis. Dat is wat de foto wil tonen en dat noemt Barthes dus het studium, het bewuste onderwerp.

Maar dan zie je haar schoenen. Een veter zit los, en de schoenen zijn net iets te groot. Niemand heeft daarop gelet toen de foto werd gemaakt. Toch prikt dat detail jou. Het raakt je, misschien omdat je er plots een heel leven in ziet: haast, armoede, ijdelheid, of gewoon een ochtend waarop ze niet goed had opgelet.

Dát is het punctum.

Nu het cruciale punt: waarom werkt dat zo? Omdat jij weet, en dit is bijna fysiek, dat die schoenen er echt waren. Dat die vrouw echt op die grond stond. Het licht viel werkelijk op haar voeten en werd vastgelegd. De foto is als een vingerafdruk van de werkelijkheid. Er is een directe, fysieke verbinding tussen dat moment en het beeld dat jij nu ziet.

Dat is wat Barthes bedoelt met emanatie: het beeld straalt iets uit van het ding zelf, zoals een gloeiend voorwerp warmte uitstraalt. Niet symbolisch, maar bijna letterlijk.

En precies daarom, zo redeneert Hoffman, kan een AI-beeld dit niet. Als je in een AI-beeld een schoen met een losse veter ziet, waren die schoenen er nooit. Er was geen ochtend, geen vrouw, geen haast. De machine heeft pixels gecombineerd op basis van patronen. Er is niets om naar terug te keren. De draad naar de werkelijkheid bestaat simpelweg niet.

Het punctum trekt aan die draad. Als er geen draad is, is er niets om aan te trekken.


Op het eerste gezicht is Hoffman’s betoog overtuigend. Dat het punctum een ontologische band vereist tussen beeld en werkelijkheid, dat is geen retorische truc, maar een serieuze filosofische observatie. En hoe graag ik Hoffman’s verhaal ook wil geloven, wringt voor mij de spreekwoordelijke schoen toch ergens.

Zo is zijn redenering over AI circulair: AI-beelden hebben geen punctum omdat ze niet menselijk zijn, en ze zijn niet menselijk omdat ze geen punctum kunnen hebben. Bovendien bewijst hij eigenlijk niet waarom een onverwacht detail in een AI-beeld de kijker niet evengoed kan treffen. Hij postuleert het.

En dan word ik als voormalig onderzoeker toch een beetje getriggerd om dit verder te onderzoeken met op zijn minst een steekproefje.

Allereerst, kan AI een punctum überhaupt herkennen?

Het antwoord lijkt bevestigend.

The punctum in mijn foto Querencia, zoals aangewezen door Claude

Deze week maakte ik in mijn studio deze foto, getiteld ‘Querencia’. Het is een Spaans concept, wat zoiets betekent als “een plek waar je je diep thuis voelt, waar je kracht uit put, waar je jezelf bent. Niet zomaar een favoriete plek, maar een plek die je ziel voedt en waar je altijd naar terugverlangt”. Ik voerde het beeld aan ChatGPT, Claude, en Gemini. Alle drie wezen ze het volgende punctum aan (aangegeven met de pijl), waarbij Claude het als volgt wist te onderbouwen:

“Voor Barthes is het punctum datgene wat prikt — niet wat het beeld wil tonen, maar wat jou treft zonder dat je erom vraagt.

In dit beeld zit het punctum voor mij niet in de bloem zelf — die is te bewust gecomponeerd, te studium. Het zit in de schaduw.

De schaduw onder het vogelnest heeft iets onverwachts: hij valt niet weg als decoratief element, maar lijkt het nest te verdubbelen als een negatief van zichzelf — als een echo, of een graf. Het object staat erboven, los, bijna zwevend. Maar de schaduw trekt het terug naar aarde.

Daarin zit iets van wat Barthes beschreef als ça a été — dit is geweest. De droge, dode bloeiwijze (een uitgebloeid allium, vermoed ik) nést in iets wat ook een nest of een ruïne is. Schoonheid die zichzelf al heeft overleefd.

Het punctum is dus voor mij: de combinatie van zweven en vallen tegelijk — het object komt los van de grond, maar zijn schaduw laat het niet gaan.”


Nogmaals, het punctum is per definitie subjectief en is wellicht anders voor de kijker dan voor de maker, maar wat mij betreft is dat punctum opmerkelijk goed gezien en verwoord.

Dan een tweede test.

Kan AI zelf beeld genereren met een duidelijk punctum?

Gemini 3/nano banana lijkt op dit moment het beste model om beelden te genereren. Die heb ik daarom gevraagd een beeld te maken met een duidelijk punctum:

Door Gemini gegenereerd beeld

Gemini wees deze drie locaties in het beeld aan als mogelijke punctum:

1. De leesbril die aan een koordje hangt

Voor mij zit het punctum in de bril die laag op de borst van de vrouw rust. Terwijl het studium van de foto gaat over ouderdom, eenzaamheid en nostalgie, is de bril een heel specifiek, bijna pijnlijk tastbaar detail. Het suggereert een onderbreking: ze was aan het lezen of aan het handwerken, maar keek op. Het is een object dat haar fysieke aanwezigheid en haar "zijn in de tijd" benadrukt. Het is geen "mooi" onderdeel van de compositie, maar een menselijk, licht onhandig detail dat de foto uit het puur esthetische trekt.

2. De lichte slijtage op de armleuning

Een ander punctum zou de versleten stof op de rand van de fauteuil kunnen zijn. Het vertelt een verhaal van jarenlang op precies dezelfde plek zitten. Het is een detail dat getuigt van de onverbiddelijke tijd (Chronos), wat Barthes vaak associeerde met de dood en het "dat-wat-geweest-is".

3. De houding van de handen

Kijk naar de manier waarop haar handen in elkaar grijpen, niet krachtig, maar rustend. De lichte glinstering van een trouwring die net zichtbaar is, kan voor een toeschouwer fungeren als het punctum: een klein detail dat de hele emotionele lading van het beeld (verlies, trouw, overleven) plotseling scherpstelt.


Tja, die trouwring. Dat is ‘m wat mij betreft.

Met dit eigen steekproefje kan ik natuurlijk nog niet concluderen dat we als creatief professionals allemaal overbodig gemaakt zijn, maar ik ben helaas toch ook nog niet helemaal gerustgesteld door Hoffman’s essay.

Misschien de belangrijkste resterende vraag is of het punctum in het beeld zit of in de kijker. Mijn steekproef suggereert het laatste.

Sander Martens

Sander Martens is a Dutch freelance photographer specialised in portrait and fine-art photography.

http://www.sandermartens.com
Next
Next

Point of reference