Protect the collection before reviewing it
Duplicate cleanup is a destructive task when the final step is deletion. Before using any detection tool, identify the working library and create an independent backup. Confirm that you can open files from that backup. A second folder on the same failing drive does not protect against drive failure, and a synchronized deletion may reach your cloud copy.
Start with a copied sample folder, not your only library. Check whether the application manages files internally. Moving photos outside Apple Photos or a Lightroom Classic catalog can break links or separate image files from editing instructions. Use supported library tools when working inside those systems.
Understand what an exact match means
An exact duplicate contains the same sequence of bytes as another file. A tool can group candidates by size and compare full-file cryptographic hashes such as SHA-256. Matching filenames alone are not evidence: cameras reuse names, and different exports may share a name. File size alone is also insufficient.
Matching full-file hashes provide extremely strong practical evidence of identical contents; a byte-for-byte comparison can provide direct confirmation. Neither method tells you which folder relationship, album membership or sidecar should be retained. Two identical JPEG files may sit beside different caption files or belong to different projects.
Changing embedded metadata changes the file's bytes. Two visually identical photos with different captions or timestamps may therefore have different hashes. This is expected and is one reason to keep exact-file detection separate from visual similarity review.
Treat similarity as a suggestion
Perceptual hashing and image-recognition methods compare visual features rather than entire file contents. They can group a resized export with its source or identify neighboring burst frames. Their thresholds trade missed matches against false positives. A high similarity score does not establish that one file is unnecessary.
Review candidates at useful magnification. Look for sharper focus, an unblurred subject, different expressions, a changed crop and editing work. Compare pixel dimensions, format, color profile and capture time. A smaller JPEG may be a deliberate delivery file, while a RAW image may retain editing flexibility that the JPEG does not.
Use a decision table, not an auto-delete rule
Write your retention rule before reviewing a large group. For example: keep the verified original in the documented master location, preserve intentional edits, and retain the associated sidecars. Do not automatically favor the largest file; an upscaled copy can be larger without containing more genuine detail.
| Candidate type | Check first | Cautious action |
|---|---|---|
| Byte-identical files | Master location, links and companion files | Keep a documented master; quarantine redundant working copies. |
| Same scene, different size | Dimensions, crop, export purpose and metadata | Keep the source and useful delivery versions. |
| RAW and JPEG pair | Editing workflow and camera rendering | Do not classify as exact duplicates. |
| Burst or near-match | Focus, movement and meaningful differences | Review visually; keep uncertain candidates. |
| Edited and unedited | Adjustments, layers and sidecars | Preserve both when the edit has value. |
Make removal a reversible stage
For ordinary file folders, move reviewed redundant copies to a dated quarantine folder outside the active import path. Keep a log of original paths and the reason for each decision. If your application has a managed trash or rejected-items workflow, understand its retention and synchronization behavior instead of improvising a file-system move.
Leave enough time to notice missing references and to complete another backup cycle. There is no universal safe waiting period: choose one that fits your restore process. Before permanent removal, verify that retained masters open correctly and that the quarantine does not contain unique sidecars, videos or motion-photo companions.
Important distinction: redundant working copies and independent backups serve different purposes. Never deduplicate away the separate copies that make recovery possible.
Verify the result and stop at uncertainty
After each batch, compare counts with your review log and open a representative selection of retained images. Test albums or projects that referenced the removed files. Confirm that your backup still contains an intact, recoverable version from before cleanup.
If a candidate's role is unclear, leave it in a labeled review group. A few uncertain files are a manageable cost. Losing a meaningful edit or the only full-resolution original is not. Once the process is predictable on small batches, you can repeat it without handing every decision to a detector.