Will my date notice my photos were AI?

6 August 2026

This gets asked more than almost anything else about generated dating photos, and it is answered almost nowhere, because the honest answer is bad for anyone selling them. So here is the honest answer, including the parts that cost us something.

The promise the category makes, and why we will not make it

Look at any product page in this category and you will find a version of the same line: photos that pass authenticity tests, results no detector can flag, images indistinguishable from the real thing. Read that carefully. It is not a claim that the photo is accurate. It is a claim that nobody will be able to tell.

That is a concealment promise, and it answers a question you should not be asking. We are not going to make it. We do not know what your date will notice, we cannot know, and neither can anyone who says otherwise in a headline.

What people are actually afraid of

Almost nobody has been dumped for using software. The fear sitting behind this question is older and more specific than that. It is walking into a bar and watching a small recalculation happen on somebody's face, the one that means you are not the person in the pictures. Everything after that is polite.

And that failure predates AI by about two decades of online dating. It is the photo from four years and one haircut ago. It is the one from before your body changed. It is the one processed until the face in it is a narrower, smoother version of the face that turns up. It is the group photo picked because it happened to catch the one angle you like of yourself. Not one of those was generated. Every one of them produces the exact moment you are worried about.

So the question is the wrong one

The useful question is not "will they notice it was AI". It is "is this a current likeness of me". Make that substitution and most of the confusion clears, because the second question has an answer you can check, and you are the only person in the world who can check it.

It also explains why a photo can be entirely real and still fail, and why a generated image can be accurate and pass without incident. The method is not what is being tested.

What actually gives a generated image away

The specific question still deserves a specific answer, so here is one. Failures cluster in the places where a model has to invent structure rather than reproduce it.

Hands and fingers. Count, length, and the way a hand meets an object it is supposed to be holding.
Teeth. Too many, too even, or with boundaries between them that do not resolve when you actually look.
Jewellery and watch faces. A watch is a small dense object with rules. Numerals out of order, hands that do not meet at the centre, a ring that changes width as it passes behind a finger.
Text. Writing on clothing, signs, book spines, labels. It comes out letter-shaped rather than as words.
Glasses. Frames that merge into the skin, arms that never arrive at the ear, lenses that fail to shift what is behind them.
Ears. Shape that drifts between images, earrings attached to nothing, an ear that dissolves into hair.
Hair against a busy background. Individual strands are where the boundary between a person and a scene tends to break down.
Repeated patterns in fabric. Stripes and checks that go out of register across a fold or a seam.
Backgrounds that do not survive a second look. Fine at a glance, then a chair with the wrong number of legs, a doorway that leads nowhere, a reflection of something that is not present.
Skin that is too even. No pores, no texture, none of the small asymmetries every real face has.

Then the one that is not about any single image. A set of six with the same light, the same distance, the same expression and the same tilt of the head reads as synthetic even when no individual picture has a flaw in it. Real photos of one person are inconsistent, because life is. This is the tell most people fail, and it is the one a better generator does not fix.

What does not give it away

The fact of generation, by itself. There is no visible property shared by all generated images that a human being can see. When somebody says they can always tell, what they mean is that they can tell when it is bad, which is a much smaller claim and a true one.

Nor does somebody simply knowing. A person who knows an image was generated and can see that it looks like you does not have a problem to solve. What creates the problem is a gap between the image and you, not the software that produced it. We wrote separately about whether and when to say anything, because the timing turns out to matter more than the fact.

That list has a shelf life

Every item above has been described as an unsolvable problem and then stopped being one. Hands were the famous tell for two years and largely are not now. Text is improving. Skin texture is improving. A list of tells is a snapshot of what current models happen to be bad at, and reading it in a year will feel like reading a phone review from 2019. Treat it as dated the day it was written, because it is.

That cuts in an uncomfortable direction. As the tells disappear, the only thing left holding an image honest is whether it is true.

The check that always runs

Here is the part nobody in this category will print. If the image is not a current likeness of you, none of the tells matter, because the check that decides this is not a detector, a browser extension, or a thread of people squinting at your ears. It is the person across the table, who has your face in front of them and your profile on their phone, and who runs the comparison without deciding to.

That check has complete coverage. It cannot be optimised against, it does not care how the file was produced, and it is the only test attached to a consequence you actually mind about. Every other test is a proxy for it. A photo that would fool a detector and not fool a human being at a table has passed the wrong exam.

What has to be true for the answer to be no

It is you. Your face, your bone structure, your body, your age. Not a quietly corrected version that would need explaining.
It is you now. Current weight, current hair, glasses if you wear glasses, beard if you have one. A genuine photograph of you from 2019 is not a current likeness either, which is the point.
You would be relaxed if they knew how it was made. Not delighted, not eager to raise it, just relaxed. If the honest answer is that you would rather they never found out, the image is doing work the rest of you cannot support in person.

If any of the three is false, a better generator does not fix it and you should not buy one, ours included. What fixes it is a current photo, however ordinary. An ordinary accurate picture beats a spectacular inaccurate one at the only moment that counts, which is the moment you walk in and they look up.

If you are not sure which of your photos still look like you

That judgement is yours and it cannot be delegated. A tool can tell you whether an image is legible: whether they can see your face, whether the light and focus hold up, whether your eyes are visible. Our free photo check does that. It returns where the photo goes on your profile, six ratings, and what to improve. Every photo is judged on its own, never against your others. It rates the photograph, not the person, and it deliberately refuses to tell you whether a photo looks like you, because it has never met you. That part stays with the only person holding the information.

And if a profile is not working at all, the mechanical causes are worth ruling out before any of this, because most of them have nothing to do with AI. Those are in what is actually going wrong on Tinder.

Written by Fottory, which generates dating photos from photos of your own face. The first two are free, and none of the advice above depends on using it.