Compare several images to find pairs that look alike — handy for spotting duplicate photos in a batch.
Your images are never sent to a server (processed in your browser)
Choose images (2 or more)
Similar pairs
How to use
Choose 2 or more images to compare.
Every pair is compared automatically and listed from most to least similar.
Pairs marked "Nearly identical" or "Very similar" are likely duplicates.
How "perceptual hashing" compares images
This tool shrinks each image down to a tiny 9×8 grid, compares the brightness of neighboring pixels, and turns that into a 64-bit fingerprint (a hash). Comparing how many bits differ between two images' fingerprints gives a rough measure of how visually alike they are. This approach is robust to resizing, mild brightness changes, and light compression — but it doesn't understand image content the way AI does.
It's well suited to catching duplicates — the same photo saved twice, or several shots from one burst — but it can't tell whether two photos show the same specific person or the same kind of object.
FAQ
Are my images uploaded to a server?
No. Hashing and comparison happen entirely in your browser.
Can it detect a cropped or heavily rotated duplicate?
Not reliably. Since this compares the overall brightness pattern of the whole image, a heavily cropped or 90°-rotated version of the same photo may register as unrelated even though the content is the same.
It says "Nearly identical" but the photos look different to me.
Since only the brightness pattern is compared, two different photos with similar tone and composition can register as similar. Treat this as a rough guide, not a certainty.
Two completely unrelated images still show 20-30% similarity. Why?
This method scores similarity by comparing 64 bits of brightness-pattern data, and even unrelated images can share some of that pattern by chance. If neither the "Nearly identical" nor "Very similar" badge is shown, treat the percentage as statistical noise rather than a meaningful match — the badge label is the more reliable signal, not the raw percentage.
How is this different from the Screenshot Comparison tool?
Screenshot Comparison strictly checks which individual pixels differ between two same-sized images — good for verifying a design change. This tool instead looks for photos that are broadly alike overall, even if sizes or details differ — good for cleaning up duplicate photos. They serve different purposes.
How many images can I compare at once?
More images mean more pairs to check (n images give n×(n-1)÷2 pairs), so processing time grows accordingly. A few dozen images is a reasonable practical limit.