"Is this photo real?" deserves an answer you can show to someone else.
Uthentic analyses a photograph or a text and returns a report: what was observed, where, how confidently, and with which limits. The score exists, but on its own it decides nothing. This page explains what actually happens inside an analysis, and what kind of document you get at the end — the honest work of an AI image detector, not a magic number.
Every report explains in words why an image is or isn't credible.
The score is still there, but it comes last: first you read what was observed in the light, the shadows, the reflections and the material. A percentage with no reasoning behind it can't be argued or defended; an explanation can.
Light, optics, depth of field and how materials behave are checked the way they would be on set.
Generated images almost always get the same things wrong: a light direction that doesn't match the shadows, reflections that don't repeat the surrounding scene, skin that doesn't react correctly to backlight. These are trade checks, not just statistics.
Suspicious areas are highlighted on the photo itself, one by one, with the reason attached.
That way you can check the reasoning yourself: if a flag doesn't convince you, you see it straight away. No flagged area without an explanation next to it.
A real photograph worked on in post-production is not a fake photograph.
This is the distinction that matters most to anyone who works with images: heavier contrast, smoothed skin, pushed colour or a replaced sky are described as edits made to the photograph, not as proof that it was generated.
File-level technical data, provenance traces and signs of local editing all go into the report.
When a file has kept its history, that history is read and reported. When it has been stripped away, the report says so instead of ignoring it.
Two distinct analyses, with two distinct reports.
For a text the question isn't only whether it feels artificial, but why: how regular the rhythm is, how predictable the word choices are, which recurring phrasing shows up. Each side stays independent, without blending the two verdicts together.
When the signals contradict each other, the report declares insufficient evidence.
Forcing a percentage on an image that is barely readable, heavily compressed or too small is the fastest way to produce a wrong accusation. We'd rather leave the verdict open and honest.
A PDF with the summary, the flagged areas, the checklist and a verification code.
It's built for anyone who needs to show the result to a newsroom, a client, a lawyer or a platform: the document stays readable even outside the app.
The file is only used while the analysis runs, then it's deleted.
It never feeds the training of any model and is never kept in an archive. That's a working condition, not a marketing promise.
A single image is never judged from one clue alone. Uthentic combines independent readings — statistical, visual, technical — and cross-checks them before writing a verdict. If they agree, the report says so clearly. If they contradict each other, that disagreement is shown instead of being smoothed into an average. A single reading, no matter how good, can't carry a judgement that someone else will use to decide something.
Every change to how those readings are combined goes through a repeatable test bench of 10,179 tests, with a category dedicated specifically to authentic photographs worked on in post-production. The goal is not to say "AI" as often as possible, but to make sure the verdict holds up alongside its own explanation.
No tool is 100% reliable, and we don't claim ours is either. What we promise is to show you the basis on which the report was written, so the final judgement stays yours.