About Retrace

Retrace answers one question about a picture: where did it come from? Give it a photo, a GIF or a short video, and it looks for the same image elsewhere on the open web — earlier copies, the pages that carry them, when they first appeared, and what can be read from the frame itself. The answer comes back as a report in which every claim carries the link it rests on.

What it is for

A picture arrives with a story attached, and there is no obvious way to tell whether the story is true. Journalists, researchers, moderators, buyers and ordinary readers all meet that moment. Retrace is a verification tool, not an oracle: it does not tell you what to believe. It shows you what is publicly known about the image, and where you can check it for yourself.

Open sources only

Everything in a report comes from information that is already public: pages anyone can open, openly available archives of the web, and what the file itself carries with it. We do not use private databases, we do not buy data about people, and we do not go behind logins, paywalls or any other access control. A finding we cannot show you as a link you can open yourself does not belong in a report.

Our own model, on our own machines

The reading of a frame — what is in it, what is written on a sign, what kind of place it is — is done by our own language model, running on our own servers. Submitted files are not handed to outside AI providers, and nothing anyone submits is used to train a model, ours or anybody else's.

What Retrace deliberately does not do

• No face search. Retrace cannot find a person by their face, and it is not built to. It answers questions about a picture, not about the human being in it.
• No biometric database. We do not create, store or compare biometric templates, and we do not build or extend any collection that could be used to recognise someone.
• No naming of private individuals. A name appears in a report only when independent public sources already publish it together with this very image, when there is a page that both carries the picture and presents that person, and when the role in question is a public one. Even then it is offered as a possibility, with the link beside it. No link, no name.
• No surveillance and no dossiers. Retrace is not a tool for monitoring, profiling or locating anyone, and using it that way is a breach of our Terms of Service.

Findings are hypotheses, not verdicts

A report keeps apart what was measured, what several independent sources agree on, and what rests on a single source with nothing to confirm it — and it says which is which, next to each claim. The open web is incomplete and archives have gaps, so a finding may be wrong and the absence of a finding proves nothing at all. Read a report as a starting point for your own checking, not as a conclusion, and never as grounds for a decision about a person.

If a report concerns you

If you are depicted in an image that someone has checked, or you hold rights to one, you can ask us to take the report down. How to do that, and what else you may ask of us, is set out in our Privacy Policy and Terms of Service.

Who builds it

Retrace is made by the Archivarix team, known for tools that recover and work with archived web content. You can find our other projects at archivarix.com, and reach us through our contact page.