Duplicate Photo Finder

About the Duplicate Photo Finder

Pick a pile of photos — or a whole folder — and find the copies: identical files (byte for byte, by SHA-256) and look-alike copies that were resized, recompressed, lightly edited or sent through a chat app, found with two perceptual hashes. Each group suggests the copy to keep — the one with the most pixels, then the sharpest, then the least compressed — and you can change the choice.

Export the list as a CSV for your records. In Chrome, Edge or Opera on a computer, a folder you choose can also be tidied directly: move the duplicates into a separate folder (the safe choice) or delete them. Your photos are analysed on your device and never uploaded.

How the photos are compared

• Identical files: only files that share their exact size with another file can be identical, so only those are fingerprinted with SHA-256 (FIPS 180-4). Equal fingerprints mean equal bytes.
• dHash (Neal Krawetz, “Kind of Like That”): the upright picture is reduced to 9 × 8 grey pixels; each of the 64 bits says whether a pixel is darker than its right-hand neighbour.
• pHash, the DCT-based hash from Christoph Zauner’s 2010 thesis Implementation and Benchmarking of Perceptual Image Hash Functions: 32 × 32 grey pixels, a 2-D discrete cosine transform, and one bit per coefficient of the 8 × 8 lowest frequencies, set when it is above the median of the 63 that are not the DC term (the bit rule of the current pHash library). The picture is reduced by area averaging rather than pHash’s blur-and-resize, so a few bits can differ from the pHash program’s own hash.
• Two photos count as similar when the pHash differs in at most the chosen number of bits (Strict 4, Normal 10, Loose 16 of 64), the dHash in at most 6 more, and an 8 × 8 colour thumbnail is close — the colour check keeps apart flat or dark pictures whose hashes are almost empty. Krawetz describes distances of 1–10 as “potentially a variation” and more than 10 as “likely a different image”.
• Every photo in a group is identical or similar to the copy suggested for keeping itself. Burst shots or a slow pan, where each frame resembles the next, therefore become several small groups, not one long chain of “duplicates” of a picture that looks different.
• Pictures are compared upright (after the EXIF orientation is applied), so a copy saved already rotated still matches.

Which copy is suggested

The suggestion follows a fixed order: most pixels (the original is usually larger than shared or thumbnail copies), then the sharpest (more fine detail, measured as the variance of the Laplacian at a common size), then the largest file (less compression), then the oldest modification date, then the shortest path. The reason is shown on every group. You decide what happens — nothing is moved or deleted until you ask.

Moving and deleting safely

Moving and deleting need the File System Access API, available in Chrome, Edge and Opera on a computer: choose the folder with Choose a folder and allow editing when the browser asks. Move duplicates puts every copy you did not keep into a folder called “Duplicates (MySmartCoPilot)” inside the chosen folder, renames clashes (“IMG_1 (2).jpg”) and saves a CSV log there with each file’s original location, so you can look through them before deleting them yourself. Delete duplicates removes files permanently — they do not go to the Recycle Bin or Trash. Both act on the groups the Show setting lists (all, identical or similar), including groups further down that are not displayed yet; the confirmation says how many files and groups are affected. In other browsers, export the CSV and tidy up in your file manager.

Limitations

• Up to 5,000 photos per scan. Very large libraries take a while: every photo has to be decoded once.
• HEIC photos can be compared only in browsers that open them (Safari); elsewhere — and for RAW files — only identical copies are found.
• Rotated or mirrored copies (where the pixels themselves were turned), heavy crops, strong filters and collages are usually not recognised as similar.
• Very similar but different photos — burst shots, or the same view taken twice — can be grouped at the Loose setting. Check each group before moving or deleting.
• Moving and deleting work only for a folder chosen with Choose a folder in Chrome, Edge or Opera on a computer. Deleted files cannot be recovered from the browser.

Privacy

Everything happens in your browser. What you enter or open here is not uploaded or stored by MySmartCoPilot.