Can people tell an AI headshot from a real one?

At the size a profile photo is normally seen, usually not. At full size, someone looking carefully can often find something — most often in the glasses, the hair edges, or skin that's been smoothed until it has no texture. Which is why it's worth checking your own at 100% before using it.

Why do AI headshots sometimes look plastic?

Because the skin texture has been smoothed away. Pores and fine lines are what make skin read as skin, and without them a face looks like a model of a face. Heavily filtered reference photos make this much more likely, since the smoothing is in what you uploaded.

Will it still look like me?

That depends mostly on your uploads. Varied, recent, unfiltered photos anchor the result to your actual face; a thin or filtered set lets it drift toward a generic average. The free preview exists so you can check this on your own face before paying anything.

Do AI headshots work if I wear glasses?

Usually, but glasses are the hardest single element to get right, so check the result carefully — frame thickness, where the arms meet the hinge, and whether reflections match the rest of the lighting. Upload plenty of photos wearing them, and judge the preview before you buy.

Buying decisions guide

Do AI headshots actually look real?

At the size a profile photo is normally seen, usually yes. Whether it looks like a photograph and whether it looks like you are two different questions, and the second is the harder one. The failures are consistent and easy to learn: smoothed skin, glasses that don't survive, mismatched details, and a face that's almost yours.

By Max Werner, who builds PoseLabAugust 16, 20267 min read

In short

Looking like a photograph and looking like you are separate questions — the second is harder and matters more.
The reliable tells are smoothed skin, glasses, mismatched left-right details, hair edges, and teeth.
At thumbnail size a good result is essentially indistinguishable; at full size and in print, judge carefully.
The best test is showing it to someone who knows you and watching whether they hesitate.

Two different questions

“Does it look real” collapses two things that are worth separating. The first is whether the image reads as a photograph rather than an illustration — whether the light behaves, whether the fabric looks like fabric, whether the depth of field is plausible. Current image tools are good at this, and for a head-and-shoulders portrait against a plain background they're very good at it.

The second is whether it looks like you, and that one is much harder and matters far more. A flawless photograph of a person who is nearly you is a worse outcome than a slightly rough photograph that's unmistakably you. Almost everything below is about the second question.

The tells, roughly in the order you'll notice them

These are the specific places where AI headshots go wrong. None of them is universal, and a good result has none of them, but this is the list to check any image against — ours or anyone else's:

Skin with no texture. Pores, fine lines, and slight unevenness are what make skin read as skin. When they're gone the face looks like a wax model, and this is the most common tell by a distance.
Glasses. Frames are the single hardest thing here — arms that don't line up with the hinge, a frame that changes thickness across the face, or reflections that don't match the lighting anywhere else in the picture.
Asymmetric details that should match. Earrings that differ between ears, a collar that's a different width on each side, buttons that don't sit on a straight line.
Hair at the edges. Where hair meets the background, look for strands that end abruptly, halo, or dissolve into the backdrop.
Teeth. Too even, too white, or slightly too many — a smile is worth zooming into before you commit to an image.
Text and logos on clothing, which reliably turn into something that looks like writing without being writing. This is why a plain top is the advice everywhere.
Ears and jaw edges, which are where the shape of a face quietly drifts away from the reference.

Identity drift is the failure that costs you

The tells above are cosmetic and mostly fixable. The one that isn't is subtler: a result that's technically clean but has averaged your face towards a more generic one. A slightly different nose. A jaw that's been quietly narrowed. Cheekbones that are a shade more defined than yours. Individually none of these registers; together they produce a photo that people who know you will describe as “nice, but it doesn't really look like you”.

This is almost always an input problem rather than a software one. Reference photos that are few, dim, filtered, or all from one moment give less to anchor to, and the result drifts toward an average. The guide to reference selfies is the fix, and it's the highest-leverage ten minutes in the whole process.

Where they hold up, and where they don't

Size decides most of it. At the size a profile photo is actually seen — a small circle in a feed, a thumbnail in a candidate list, a face on a team page — a good result is essentially indistinguishable from a studio photograph, because none of the tells above survive at that scale either. That happens to be where the overwhelming majority of profile photos live.

Zoom to full size and it changes. Someone looking carefully, on a large screen, with a reason to be suspicious, can often find something — most likely in the glasses, the hair edges, or the skin. And print is the least forgiving surface of all. If your photo is going on a conference banner or a printed page at any size, judge it at 100% first rather than at the size it appears on your phone.

What we don't attempt

It's easier to trust the answer on the things we do well if we're clear about the things we don't do at all. We create head-and-shoulders portraits. There are no full-body images, no groups, no photos in your actual office, and nothing involving hands — hands remain a well-known weak point for image tools generally, and the honest response to that is to frame them out rather than to hope.

The other limit worth stating plainly is that we can't make a result that's better than its references. A set of dim, filtered, two-year-old photos produces a mediocre gallery, and no amount of paying more changes that. A photographer controls the input; here, you do.

How to judge it without trusting us

Run four checks, in this order. Look at the image at full size and go down the tells above. Look at it as a small circle on your phone, since that's the size that actually matters. Then do the test that beats both: show it to someone who knows you well, without saying what it is, and watch whether they pause. People are extremely good at this about faces they know, and far better than you are about your own.

The fourth check is time. Look at it again the next day. Most images that feel slightly wrong on first viewing either resolve into fine or turn out to be the ones you never end up using.

Do all of that before you pay

This is why the preview is free rather than discounted. You upload your own selfies, get one finished headshot back in about a minute with no card and no account, and run every check above on your own face instead of on the samples we chose to publish. Nothing about our marketing is a better answer to “does it look real” than an image of you.

If it doesn't hold up, don't pay. That isn't a rhetorical flourish — the honest failure rate on this is not zero, it's usually traceable to the uploads, and a preview that misses is information rather than a loss. See the finished styles at full size as well, since those are the outputs the process is aiming at.