What an image checker can and cannot tell you

The three categories

Measured. The file says so, and code that reads it cannot be wrong: 1,600 by 1,200; 840 KB; JPEG; 4:3; no alpha channel; GPS coordinates present. These are facts, and a tool that shows you the number rather than a verdict lets you check it yourself.

Observed. Somebody or something looked at the picture and formed a view: the background is uneven, the product is cropped, the face is off-centre. A careful human is good at this and a vision model is sometimes good at it. Neither is certain, and a page that presents an opinion in the same voice as a measurement is misleading you about which you have.

Not answerable. Whether a passport office will accept this photograph. Whether a marketplace will rank this listing. Whether a buyer will click. No amount of examining the file gets you there, because the answer lives in somebody else's decision.

Why the distinction matters when you act

A measured failure is worth fixing immediately: it is certain, and it is usually a resize or a compression away. An observed problem is worth looking at yourself before you act, because the observer may be wrong. And a claim in the third category is worth ignoring entirely.

This is also why a single compliance score is a bad idea. Averaging forty missing pixels against a shadow on the wall produces a number that hides the only thing you needed to know: one is a minute's work and the other is another photo session.

What to look for in a tool

Three habits separate a checker worth trusting from one that just looks confident:

1. 1It shows the measurement, not only the verdict. "1,200 px - at least 1,600 required" beats "failed".
2. 2It says where a requirement came from. A figure published by the authority and a figure repeated by other websites are different things, and a page that grades them the same is telling you less than it appears to.
3. 3It admits what it did not check. A checker that silently passes everything it cannot measure is the most dangerous kind, because the reader assumes the silence means approval.

Where AI fits, and where it does not

A vision model is genuinely useful in the second category. Asked whether a background looks plain and evenly lit, or whether anything covers the face, it can offer a reasonable opinion - and that is exactly the half a measurement cannot reach.

It is not useful in the third. A model asked whether a photograph will be accepted will answer, fluently and with no standing to. The safeguard is to ask it what it can see rather than what it thinks the verdict is, and to treat a low-confidence answer as unassessed rather than as a pass. An honest "could not assess this" is worth more than a confident guess, because the cost of a wrong pass here is a rejected application.

Frequently asked questions

Can a tool tell me if my image will be accepted?

No, and be wary of anything that says it can. A tool can tell you whether the file matches published figures, and it can relay an opinion about the picture. Acceptance is a decision made by whoever receives it, against criteria they may not publish in full.

Why does a checker say something cannot be checked?

Because a rule exists that the file cannot answer - the background colour of a photograph it could not sample, or whether the item pictured is the one being sold. Saying so is more useful than quietly passing it, which would leave you believing it had been examined.

Is my image uploaded when it is checked?

Not for the measured half - that happens in your browser and produces no file at all. An AI assessment is different by nature: it has to send a copy to the provider you chose, and it says exactly what it will send before it sends anything.