rumbled. Reads the file and the pixels. Shows its working.

Rumble the fakes

Someone passing an AI image off as a real person? Drop it in. Rumbled reads what is hidden in the file and looks hard at the pixels, then tells you what it found and which tool likely made it.

Drop an image here
or tap to choose, or paste from your clipboard
JPGPNGWEBPScreenshots welcome

The file check runs on your device the instant you drop an image, and uploads nothing. The access code unlocks the deep pixel scan, which reads the picture itself and works on screenshots.

Which images work best
1 / BESTThe original fileSent straight from the source, nothing stripped. Both checks run at full strength and the file may still carry its maker's fingerprints.
2 / GOODA screenshotScreenshots throw away the hidden file data, but the pixel check reads the picture itself, so it still works well.
3 / WEAKESTA re-saved or shared copyPassed through WhatsApp, re-saved, shrunk. Every squeeze rubs out clues, so confidence drops. Get the original if you can.

Reading the image

    The image you uploaded
    Format
    -
    Dimensions
    -
    File size
    -
    File check: done Deep scan: not run
    Reading...

    In the file On-device, no upload

      Good to know

      How to read this

      No single line is proof. It is the pile of small things together that tells the story, never any one of them on its own. A heavily retouched real photo can look suspicious, and a brand new AI tool can slip through clean.

      Your own eyes still matter most. Put a person's photos side by side and check the things that cannot change within one shoot: a piercing, a tattoo, the shape of a logo, a ring on the same finger. Look at lace and fishnets, where the machine still cannot hold a repeating weave, and at the background, the sockets, the labels, the bits nobody thinks to fake.

      Metadata beats guessing. When the file still holds its maker's mark, that is the strongest thing here. When it does not, that proves nothing on its own, because a screenshot rubs every mark away.

      What it cannot catch

      Rumbled answers one question: was this image made by AI? It cannot tell you whether a real photo has been edited by hand. A pulled-in waistline, a smoothed face, a reshaped body, those are real photos that someone has warped in an editing app, and they will still come back as "nothing found" here.

      For those, your eyes are the tool. Warping bends everything around it, so look at the background right next to the part that looks too perfect: door frames that curve, tiles that ripple, a straight edge that suddenly is not. And notice how often these shots use a plain, empty background, that is not an accident.

      About SynthID

      SynthID is Google's invisible watermark, and it only ever turns up in images made by Google's own tools. Rumbled cannot read it directly, because Google keeps that behind its own app. If you think an image is Google made and it is clothed, upload it in the Gemini app and ask whether it was made with Google AI. Finding nothing means nothing though, most tools do not use it at all.

      Your privacy

      The file check runs entirely on your device and uploads nothing. The pixel check sends that one image to a detection service to be analysed and not stored. Do not put in anything you would not want analysed, and take screenshots of other people's images with the same care you would want for your own.