Drop a folder in and find the files that are genuinely identical. Matching is on a SHA-256 of the contents, so names, dates and folders are irrelevant — 'IMG_4021.jpg' and 'IMG_4021 (copy).jpg' match because every byte is the same, and two unrelated files with the same name do not. Files whose size nothing else shares are ruled out before a byte is read, which is what makes this fast on a real photo library: hashing is the slow part, and most files are eliminated without it. You get the sets, the wasted space, and a list you can copy. Nothing is uploaded, and nothing is deleted — a web page cannot remove files from your computer, so the decision and the deleting stay with you.
| Case | Reported? |
|---|---|
| Same file, different name | Yes |
| Same file, different folder | Yes |
| Photo re-saved at lower quality | No — the bytes differ |
| Same song at two bitrates | No |
| Different files, same name | No |
No. Every file is read and hashed inside your browser and nothing is transmitted. That is the only way this makes sense — uploading a whole photo library to find duplicates would be absurd.
No, and it cannot. A web page has no ability to delete files from your computer. You get a list of which files are duplicates and where they are; removing them is something you do yourself, after checking the paths.
Identical only. Matching is byte-for-byte, so a photo re-saved at a different quality, resized, or exported from a different app is a different file and will not be reported. Perceptual similarity is a much harder problem and a different tool.
Because files with a unique size are skipped entirely. Two files of different sizes cannot have identical content, so there is no reason to hash them. On most folders that eliminates the majority of the work before any reading happens.
With SHA-256, no — not in any practical sense. Deliberate collisions have never been produced and accidental ones are far beyond astronomically unlikely. If two files match here, they are the same file.
Each file is read one at a time rather than all at once, so the count matters more than the total size. Thousands of files is fine; a folder with a single enormous file may be limited by available memory.
Files are matched on a SHA-256 of their contents, so a match means every byte is identical — names, dates and locations are irrelevant, and a file renamed twenty times is still the same file. The performance trick that makes this usable on a real folder is that two files of different sizes cannot possibly have the same content, so every file with a size nothing else shares is ruled out before a single byte is read. On a photo library that typically removes most of the work, which matters because hashing is the slow part and everything else is bookkeeping. Hashing runs through the browser's native Web Crypto rather than JavaScript, so it moves at close to disk speed. An unreadable file is skipped rather than failing the whole scan, since one permission error part-way through a large folder should not discard the results for everything already processed. Groups are sorted by wasted space rather than file count, because ten copies of a small text file matter far less than two copies of a video. Worth being clear about the limits: this finds byte-identical files, not similar ones — a photo re-saved at a different quality, or the same song at two bitrates, are different files and will not be reported. And nothing is deleted: a web page cannot remove files from your computer, so this produces a list for you to act on.