This Person Does Not Exist: a website of endless fake faces

This Person Does Not Exist is a website that shows one large, convincing photograph of a human face, then quietly reveals that the person in it has never existed. Reload the page and a different stranger appears, just as detailed and just as invented. It is worth a look because it turns an abstract claim, that software can now conjure a believable human out of nothing, into a single image a visitor can sit with, study, and slowly stop trusting.

In brief

• This Person Does Not Exist, at thispersondoesnotexist.com, fills the screen with a portrait of a person who does not exist, generated entirely by artificial intelligence.
• Reload the page and the site produces a brand-new face, which, as Wikipedia notes, is the whole mechanic: a new face on each reload.
• It was built in February 2019 by Uber engineer Phillip Wang, using Nvidia's StyleGAN face-generating software, according to Wikipedia.
• There is no sign-up and nothing to do except refresh, look closely, and wonder how a machine learned to fake a human this well.

What This Person Does Not Exist actually is

Open thispersondoesnotexist.com and the browser fills, edge to edge, with a single human face. There is no menu, no logo and no caption, just a person looking back with the flat calm of a passport photo. The mechanic is almost aggressively simple. One portrait fills the tab, and reloading the page swaps it for another. Wikipedia, describing the project, says the website "displayed a new face on each web page reload," which is very nearly the entire feature list.

The restraint is the point. Most websites want a visitor to click, scroll or subscribe, and this one asks for nothing at all. It presents a face, lets the silence do the work, and waits. The unease arrives a beat later, once the caption in the visitor's own head, that this is a photo of a real person, turns out to be false.

Sit with it for a minute and the variety becomes its own small marvel. Reloading turns up faces of every apparent age, background and mood: a smiling child, a tired man in his fifties, someone caught mid-blink under hard studio light. None of them share a template, and none of them share an original. Each is a fresh composite, generated in the instant the page loads and gone the moment it is replaced, seen once and then lost for good.

The machine behind the faces

The portraits are not photographs and not drawings. They are the output of a system called StyleGAN. According to Wikipedia, StyleGAN is "an extension to the GAN architecture introduced by Nvidia researchers in December 2018" and was "made source available in February 2019." The underlying research was published in a paper titled A Style-Based Generator Architecture for Generative Adversarial Networks, and Nvidia later released the code itself on GitHub.

StyleGAN did not stand still. Wikipedia records that a second version, StyleGAN2, was published on 5 February 2020, and that it "removes some of the characteristic artifacts and improves the image quality." Each revision made the invented faces harder to catch, which is either reassuring or alarming depending on the day.

A system like this does not invent faces out of thin air. It learns from a vast collection of real portrait photographs, absorbing the statistical patterns of human features, lighting and skin until it can assemble new arrangements that obey the same rules. The result is not a copy of any single photo in that collection but a plausible average made specific, a face that fits every pattern of a real one while belonging to no one.

How a GAN invents a face

The G in GAN stands for generative, and the rest describes a contest. A generative adversarial network pairs two programs and sets them against each other: one network invents images, and a second network judges whether each image looks real or fake. The judge's verdicts train the inventor, round after round, until the fakes are good enough to slip past. This Person Does Not Exist is simply the inventor's finished work, framed and hung in a browser tab.

One engineer and a very fast weekend

The site has a single, human origin story. Wikipedia states that "in February 2019, Uber engineer Phillip Wang used the software to create the website This Person Does Not Exist," wiring Nvidia's released model to a web page that served a fresh face on demand. The heavy research was Nvidia's. The small, memorable idea of pointing it at the public was Wang's.

It spread almost immediately. Technology outlets picked it up within days, among them The Verge, and the site became a standard reference point in every later conversation about synthetic media. The name helped. "This person does not exist" is a full argument compressed into four words, and it travels well.

The timing mattered too. In early 2019 the word deepfake was moving from research labs into ordinary conversation, and much of that talk was abstract and hard to picture. This Person Does Not Exist made it concrete. Anyone could open one link and watch flawless fictional humans appear on demand, no expertise required, which did more to explain the stakes than most articles managed. The demonstration was the argument.

The tells, and how they faded

For a while, the faces could be caught. Wikipedia notes that early StyleGAN images carried "characteristic artifacts," and a casual sport grew up around spotting them. Backgrounds tended to melt into abstract smears. Earrings arrived in mismatched pairs. Teeth blurred at the edges, glasses bent where the frame should have been straight, and stray hair dissolved softly into skin.

Those glitches were the seams of the illusion, and they were reassuring, because they meant a careful eye could still tell the difference. The trouble is that the seams keep closing. With StyleGAN2 tidying away many of the old artifacts, the reliable tells grow rarer, and the honest answer to "is this real?" drifts from probably not toward hard to say.

A whole family of things that do not exist

The format turned out to be easy to borrow. Once a single page could generate an endless supply of fake faces, imitators applied the same trick to almost everything else, producing sites that invent cats that do not exist, rooms that do not exist, and words, resumes and works of art that do not exist either. The phrase itself became a small genre label, a shorthand for anything a machine has quietly fabricated.

This Person Does Not Exist sits at the head of that family as the plainest, and somehow the most affecting, member. A fake armchair is a curiosity. A fake person, rendered with every pore and stray eyelash in place, lands somewhere closer to the spine. That is why the original still draws visitors long after the novelty of the format wore off. The subject is not a chair or a word but a face, the one image a person is wired to read as another human being, and that wiring does not switch off just because a caption insists the human is not real.

Why a fake face is worth a moment

Part of the appeal is the pure uncanny jolt: the second or two where a stranger's face reads as ordinary before the label catches up and rewrites it as impossible. It is a small, repeatable shiver, and reloading for another one is oddly hard to resist.

There is a right way to use the site, and it is slower than it seems. The reflex is to refresh quickly, treating the faces as a slot machine, but it works better as a staring contest. Load one face and hold on it. Try to believe the person is real, then try to find the seam that proves they are not. Then reload, once, and start again.

Used that way, a page with exactly one feature turns into something closer to a test of attention. Linger on a face and the small contradictions surface, an ear that sits a touch too high, a collar that fuses into a shoulder, a pair of eyes focused on nothing. Those flaws are strangely comforting, proof that a person really is not there.

There is a more useful side effect, as well. A minute spent watching flawless humans stream out of a browser tab is a fast education in why a photograph, on its own, is no longer proof of anything. The site never lectures. It simply shows the thing working, over and over, and lets the visitor draw the obvious and slightly unsettling conclusion about the images that fill the rest of the web.

Where it sits in the quiet web

This Person Does Not Exist belongs to the long tradition of the single-serving site: one web address, one idea, no clutter. It is the same instinct that runs through the hand-picked collection at boringwebsites.art, which gathers calm, curious corners of the web worth a few unhurried minutes.

Readers drawn to sites built around an endless, impersonal machine will recognise the shape of Every UUID, which scrolls through an unimaginably long list generated on demand. Those who prefer their machine-made art in motion can wander through CacheMonet, which loops random images into an endless collage. For a wider tour of the genre, this blog's roundup of useless websites to visit when bored maps out where a one-trick page like this one fits among its stranger neighbours.

FAQ

What is This Person Does Not Exist?

It is a free website, at thispersondoesnotexist.com, that shows a single full-screen portrait of a human face generated by artificial intelligence. Reloading the page produces a new face each time, which Wikipedia describes as the site's core behaviour. There is no account, no menu and nothing to buy: the only interaction is refreshing to summon another invented stranger.

Are the faces real people?

No. Each face is synthesised from scratch by software and does not correspond to any specific living person. The images can resemble real human beings closely, which is exactly the point, but they are assembled statistically rather than photographed, so the person on screen has no name, history or existence outside the pixels of that single image.

How does the site make the faces?

It uses StyleGAN, a face-generating system that Wikipedia says Nvidia researchers introduced in December 2018 and made source available in February 2019. StyleGAN is a generative adversarial network, meaning it trains one program to invent images and another to judge them until the results look convincingly real. The technical details appear in Nvidia's paper, A Style-Based Generator Architecture for Generative Adversarial Networks.

Who made it, and when?

According to Wikipedia, the website was created in February 2019 by Phillip Wang, then an engineer at Uber, using Nvidia's publicly released StyleGAN model. The research behind the faces came from Nvidia. Wang's contribution was the simple, viral idea of putting that model behind a single web page that hands out a new fake person on every reload.

More boring websites like this: Generative art websites.