The trend is a pair of pictures: someone today, and the same person at 60 as imagined by an image model. It is genuinely fun, it takes two minutes, and almost everything written about it stops before the two questions worth asking — how much of it is real, and what you just handed over.
The steps
- Open ChatGPT and start a new chat. Image generation works on the free tier with a daily limit, and without one on paid plans.
- Press the + button and upload a recent photo. Face the camera, in good light, no sunglasses — the model can only age what it can see.
- Give a specific instruction. This works: "Based on this photo, show me what I might look like in my forties." You can ask for several at once: "…at 40, 50 and 60, as three separate images."
- Wait a minute or two per image, then refine in the same chat — "keep the beard", "less grey" — and it will redraw.
What the model is actually doing
It has seen an enormous number of photographs of people at every age and has learned what ageing looks like on average: skin texture, hairline recession, the fall of the cheeks, the grey, the way the eyes settle. It applies that average to the geometry of your face.
It does not know your genetics, whether you smoke, how much sun you get, what you weigh, what you do for a living or what illness you will have — which is to say it does not know any of the things that actually decide how one specific person ages. The output is a plausible generic older version of you, and plausible is not predicted.
The comparison that makes this clear
Age progression is a real forensic discipline, used to search for long-missing children. How it is done is almost the opposite of what the model does.
A trained forensic artist working on a child missing for ten years does not apply average ageing. They gather photographs of the child's blood relatives at the target age — the mother at 18, the father at 18, older siblings — and build the progression from inherited structure: how the jaw developed in that family, how the nose grew, where the hairline went.
In other words, the professional method is built entirely on the information the model has no access to. The family album is the data, and the model has only your single face. That is the honest size of the gap, and it explains why a forensic progression is used as an investigative lead and still gets the face wrong often enough that practitioners say so plainly.
Why it flatters
AI-aged faces come out consistently healthier, better-rested and more symmetrical than real people of that age, and the reason is in the training data rather than in any instruction to be kind.
Photographs of older people that end up on the internet in quantity are disproportionately posed, lit, selected and retouched — professional portraits, stock photography, publicity images, the good picture out of thirty. People photographed in poor health, poor light or without a reason to be photographed are badly underrepresented. The model learns "what a 60-year-old looks like" from that filtered sample, so its average is the average of a flattering subset.
This is not a quirk of age progression. It is the general reason image models produce an idealised version of almost any category, and noticing it here is good practice for noticing it everywhere else.
What you handed over
To do this you upload a clear, well-lit, front-facing photograph of your face. That is the most useful image anyone can have of you: it is the ideal input for face matching, and the ideal seed for a face-swap or a synthetic video.
Three practical points.
Your own settings. By default the content of your chats may be used to improve the company's models. You can turn that off — in ChatGPT, under Settings → Data controls — and delete the conversation afterwards. Do that first rather than after.
Where the photo actually goes. Every viral image trend is followed within days by apps and websites promising the same effect. Those are usually not the original tool, and the ones that ask for access to your whole photo library rather than one picture are asking for something the effect does not require. If an app wants your gallery to age one face, that is the product and the picture is the pretext.
Other people's faces. Doing this with your own photograph is your decision. Doing it with a friend's, a colleague's or a child's is a decision about someone who was not asked, and a child's face especially is not yours to upload. In places where face photographs are already raw material for identity fraud and for harassment by fabricated video, that is worth a moment before the fun — not as a reason to avoid it, but as the reason to use your own picture and nobody else's.
