Reverse image search blocker
Make a photo harder to match by reverse-image search
Reverse-image engines don’t compare pictures pixel for pixel; they boil each one down to a short “fingerprint” and match on that. This tool changes the pixels just enough to change that fingerprint, with a faint warp, a tiny rescale and a little noise, so the photo still looks the same to you but reads as a different image to a hash-based search. Turn the strength up if you want a bigger change.
Runs entirely in your browser. Your photo is never uploaded or stored.
WARNING: may be slower on older machines or large photos.
Working…
This is the changed photo. Hold Compare to see the original underneath. Set any slider to 0 to turn that change off.
JPEG re-compresses on the way out, which changes the fingerprint even more; PNG keeps every pixel exact.
What it actually changes
A perceptual hash looks at the broad shape of an image, a shrunk-down, low-detail version of it, and ignores small stuff. To move that hash you have to disturb that broad shape a little without making the photo look wrong. Each slider is one way of doing that, and you can turn any of them off by setting it to 0:
- Warp bends the image with a gentle, invisible wave.
- Stretch changes the aspect a hair; because the hash squashes the photo to a tiny thumbnail, a shape change moves it more than most.
- Rotate tips the image a fraction of a degree.
- Rescale crops a sliver off the edges and scales back, so the grid the hash samples on lines up differently.
- Block jitter shifts the brightness of small tiles, right at the scale a DCT-based hash reads.
- Noise sprinkles a light grain over everything; this one dents the weaker, thumbnail-style hashes more than the strong ones.
The geometric ones, warp, stretch, rotate, rescale, do the heavy lifting against strong hashes; block jitter and noise mostly help against the simpler ones. Stack a few and the fingerprint moves well off its original value while the picture still reads the same.
What it does not promise
This is not a cloak against face recognition or against a human who looks at the picture, and it will not defeat a search engine that keeps a fuzzy, edit-tolerant index. Strong hashes are built to survive small edits, so a light setting may still match; a stronger setting changes the fingerprint more but also shows more, so pick the least that does the job for you. If a copy of the photo is already indexed somewhere, changing your copy does nothing to that one.
Free here, more behind an account
The in-browser version works on one photo at a time with a generic warp-and-noise pass. A stronger, targeted version, tuned against specific hash families and run over a batch in one go, is an account feature rather than something to do pixel by pixel here.