Someone sends you a photograph and the question is whether it is what it claims to be. TAMPER runs the standard checks: how the picture responds to being recompressed, whether any part of it is a copy of another part, whether the JPEG block structure is intact, and what the file says about the software that made it. It gives you places to look. It does not give you an answer, and anything that claims to is selling something.
Drop an image here, or click to choose one
JPEG works best. Nothing is uploaded.
The image
Error level map
Brighter means that area changed more when the picture was saved again. Read it as "look here", nothing more.
This is the part to read. Error level analysis is the most misread technique in amateur image forensics: a bright area means that part of the picture responded differently to being recompressed, and sharp edges, text, logos, a pasted region, a resized region and a dozen innocent edits all do that. A clean result does not mean a photo is untouched, because saving once more, taking a screenshot of it, or posting it through any social network erases the signal completely. A bright patch is not evidence of forgery. Cloned-block detection has the same problem in reverse: brickwork, foliage, windows and any repeating texture genuinely repeat, and the tool will say so. Everything here is a prompt for a human being to look harder at a specific spot, and none of it belongs in an argument as proof. The image is read in this tab and never uploaded.
Questions people ask
How can I tell if a photo has been edited?
There is no single test. The signals worth looking at are differences in compression error across the image, which show where an area has been through a different number of saves; duplicated blocks, which appear when part of the picture was copied to cover something up; noise that does not match between regions; and metadata naming an editor, or a modified date later than the capture.
Is error level analysis reliable?
Not on its own, and it is badly over-read. Resaving, resizing, screenshotting and passing an image through a messaging app all change compression across the whole picture and produce patterns that look like edits. Treat any result as a reason to look harder at a specific area, never as a conclusion.
Can these checks detect AI-generated images?
Not reliably. They are built to find local inconsistencies within a photograph, and a fully generated image has no original to be inconsistent with. Some generators write metadata or provenance credentials, which is a far stronger signal when present, and whose absence proves nothing at all.