The dedicated machinery
Most pictures, your brain processes in one way. A landscape. A chair. A dog. These are handled by the visual cortex as a general-purpose object recognition problem — features are extracted, compared against memory, a match is found. Efficient. Functional. Largely unconscious.
Faces are different.
Your brain has dedicated neural architecture — the fusiform face area, located in the temporal lobe — that does nothing else but process human faces. Not objects that resemble faces. Not caricatures of animals. Human faces. This region is so specialised that when it is damaged, the result is a condition called prosopagnosia: the complete inability to recognise faces, even of people known for an entire lifetime, while all other vision remains entirely intact.
The deficit is categorical. Selective. Faces only.
The brain does not treat faces as images. It treats them as a separate category of perception, governed by different rules, processed by different machinery. A face is not seen — it is identified.
What a caricature artist actually does
A photograph captures light reflected from a surface. It is accurate in the way a recording is accurate — it contains everything, weighted equally.
The brain does not weight everything equally. When you recognise a face, you are not performing a pixel-by-pixel comparison. You are responding to a small number of salient features — the specific proportions, the relationships between landmarks, the qualities that are unique to that face and deviate meaningfully from an average. Your fusiform face area has already decided which features matter. Everything else is background.
A skilled caricature artist finds those features. Not by guessing, and not by formula — by looking, long and carefully, at a specific face until the signal becomes clear. What is large relative to the norm? What proportion is unusual? What is the relationship between the eyes and the nose that makes this face this face and not any other?
Then those features are amplified, in oil paint, onto canvas.
I have been doing this for over twenty years. I estimate roughly twenty thousand hours spent looking at faces with the specific intention of understanding what makes each one unique. You cannot buy that with a subscription.
The result is something physiologically unusual: an image that the brain's face-recognition system finds easier to process than a photograph. The signal has been cleaned of noise. The identifying information has been amplified. People who see a good caricature of someone they know often report that it is more recognisable, more present, than a photograph. This is not whimsy. It is a direct consequence of how face perception works.
It also explains why a poor caricature feels so wrong. If the wrong features are exaggerated — or if features are exaggerated beyond the threshold at which the face-recognition system can still resolve them — the result registers as grotesque rather than true. The line between a caricature that captures someone and one that merely insults them is exactly the line between working with the perceptual system and working against it. Finding that line is what the twenty thousand hours is for.
Why this cannot be replicated
An AI image generator processes faces as pixel data. It has been trained on statistical relationships between features and can produce outputs that look face-like, portrait-like, even caricature-like. What it cannot do is engage with the perceptual system that makes a caricature resonate.
It has no access to what the fusiform face area actually cares about, because that information does not live in the pixels — it lives in the relationship between a specific face and the neural architecture of the people who know it. An AI has no model of what makes your father's face your father's face to you. It has only averages.
A caricature painted by a skilled artist is not a stylised photograph. It is a collaboration — between the painter's trained eye, the subject's unique face, and the perceptual machinery of everyone who will ever look at it. The painter is working in the same domain as the viewer's brain.
An algorithm is not.
"If we stop learning these things — could we ever get them back?"
The work
Each commission begins with a period of sustained looking — at photographs, at video where possible, at the face in motion and at rest. I am looking for the signal: the features that distinguish this face from every other face, the proportions that the brain of someone who knows this person will immediately recognise as true.
The painting takes as long as it takes. I do not work to speed. I work to the point where the face on the canvas is more recognisable, to the people who matter, than any photograph you could take.
If you have someone in mind — a retirement, a landmark birthday, a person who has been impossible to buy for — I would like to hear about it.
I left caricature at the top. But the twenty thousand hours don't go anywhere.
See the work