How to Remove Duplicate Photos Safely
By Jay Harb · Published · 9 min read
Deleting duplicates is easy. Deleting duplicates without losing the good copy is the part that needs a method. Here is the order to work in, and the four traps that look like duplicates and are not.
The risk is not deleting too little
Every duplicate finder will happily delete thousands of files for you. The question that matters is whether the copy it keeps is the one you wanted, and the failure mode is silent: you free up 40 GB, feel good about it, and discover two years later that the version of your wedding photos that survived is the 900 KB one a messaging app made.
So the method below is built around one principle: never let the tool decide which copy dies without you having set the rule.
There are three quite different things software calls a duplicate, and they carry different risk.
Byte-identical files. Same file, copied twice. These are found by hashing the file contents, and deleting the extra copy is completely safe — there is genuinely no information difference between them. This is the biggest category in a consolidated library and it is the one to clear first.
Visually identical files. The same photo, but re-encoded, resized, rotated, or re-saved at a different quality. Found by perceptual hashing: reducing the image to a small fingerprint of its structure, so two versions of the same picture produce nearly the same fingerprint even though not one byte matches. These are real duplicates, but which copy to keep is now a decision.
Visually similar files. Burst shots, three attempts at the same group photo, a panorama's source frames. Perceptual hashing finds these too, at a looser threshold. These are not duplicates. They are different photographs that look alike, and a tool that deletes them automatically is deleting your photos.
If your duplicate finder has a similarity threshold, this is what it controls. Start strict.
Step 2: Do the safe pass first
Run exact-match (byte-identical) removal across the whole library before touching anything fuzzy. It is risk-free, it is fast, and on a library that has been merged from several backups it typically removes the majority of the total.
Doing this first also makes the fuzzy pass much easier to review, because you are no longer wading through groups of eight identical copies to find the one real decision.
Step 3: Set the keep rule before you review anything
Decide, in advance, what "the good copy" means. A sensible default, in priority order:
- Highest resolution. Almost always the original.
- Largest file size at equal resolution. Less compression.
- Has EXIF metadata. Messaging apps and social exports strip it; originals have it. A copy with a real capture date and GPS is worth more than one without, even at the same pixel dimensions.
- Oldest file modification time. A weak signal, but a tiebreaker.
Note the trap in rule 2: bigger is not always better. A PNG screenshot of a JPEG photo is larger than the JPEG and worse. Size only breaks ties at equal resolution and equal format.
Step 4: Review by group size, largest first
Sort the duplicate groups by how many files they contain and work down. This is not just efficiency — the big groups are almost always the safe, obvious ones (a folder copied three times), so you build up momentum and confidence before reaching the ambiguous pairs.
While reviewing, look at the paths, not just the thumbnails. A group where one copy lives in Photos/2019/ and the others live in Downloads/, WhatsApp Images/ and Desktop/temp/ is telling you exactly which one is the original.
Step 5: The four things that look like duplicates and are not
Burst and continuous-shooting frames. Eight nearly-identical shots of a child running. Keep the sharpest, but this is a photographic choice, not a cleanup.
Bracketed exposures. Three frames at different exposures for an HDR merge. They look similar to a perceptual hash and are useless individually — but if you shoot brackets, deleting two thirds of them destroys the ability to re-merge.
Your own edits. An original and a cropped, colour-corrected version you actually prefer. Both are worth keeping, and which is "the original" is now the wrong question.
RAW + JPEG pairs. If your camera writes both, they are the same photograph in two formats — and the JPEG is not a duplicate of the RAW, it is a preview of it. Most duplicate finders do not group these because the files look nothing alike to a hash, but some do. Do not let one delete your RAWs.
Step 6: Delete to the recycle bin, not permanently
This is the single most important safety rule and it costs nothing. Send removals to the system trash rather than unlinking them, then do not empty it for a month. If you find something missing, it is right there.
If your tool offers "permanently delete" as the default, change it before your first run.
Step 7: Verify before you empty
After the pass, three checks:
- Photo count. Did it drop by roughly what you expected? A far bigger drop than expected means the similarity threshold was too loose.
- Spot-check a few years. Open a month you remember and confirm nothing obvious is gone.
- Check total library size. The space reclaimed should be plausible against the count removed — at 3–5 MB per photo, removing 6,000 duplicates should free somewhere in the region of 20–30 GB. An implausible number in either direction is worth investigating before emptying the bin.
Most tools have one, and used correctly it is a big time saver: it applies your keep rule (step 3) to every group at once so you only review the exceptions. Used carelessly it is how the wrong copies get deleted en masse.
The safe pattern is: auto-select, then scroll the whole selection before confirming, looking specifically for groups where the auto-selection kept something in a suspicious folder. It takes minutes, not hours, and it is the step that catches the systematic mistake rather than the individual one.
Doing it regularly is easier than doing it once
The reason people end up with 6,000 duplicates is that the job only ever gets done during a crisis. Running a scan after each big import — a Takeout export, a phone backup, an SD card dump — keeps the number in the dozens, and dozens is a five-minute review rather than an afternoon.
In Pluto Photos
The duplicate photo finder in Pluto Photos uses perceptual hashing, so it catches resized and re-encoded copies rather than only matching filenames. Groups are reviewed side by side with resolution and file size shown, removals go to the system trash, and the whole scan runs locally — no photos are uploaded to compare them.
If you are mid-consolidation, the guide to organizing a very large library covers where de-duplication fits in the order of operations, and this is what a real 6,000-duplicate cleanup looked like. Download Pluto Photos to try it — the free tier will scan a 500-photo folder, which is enough to see how the grouping behaves before you point it at everything.
Part of the Duplicate Photo Guides guide.
Download Pluto Photos for Windows, macOS or Linux, or read more from the blog.