Czkawka vs. 1,040 Files: What It Caught, and What It Missed

Every duplicate finder makes the same promise: it will find every copy hiding on your disk. Most reviews answer by repeating the feature list. We wanted results instead, so we built a trap. One folder, 1,040 files, about 11 MB, with known duplicates planted inside and a few decoys designed to fool the scanner. Then we ran Czkawka on it and checked the output against the answer key.

Czkawka is a free, open source cleaner written in Rust. It finds duplicate files, similar images, empty folders, oversized files and more, on Windows, macOS and Linux. One thing to know before downloading: the project is mid-transition. The classic GTK interface stopped at version 12.0, and the new official app is called Krokiet. Both sit on the same engine, which also exists as a command-line edition, and that CLI is what we tested, not Krokiet's interface. The engine underneath is the same one Krokiet drives.

The trap

The corpus imitates a messy drive. Picture the classic situation: folders named "Photos", "Photos copy" and "Photos FINAL", years of backups merged together, copies of copies, nothing labeled. To recreate that safely, every file in our set is synthetic. No personal data was scanned, and nothing was deleted at any point.

Planted exact duplicates: a 20 KB text report copied three times under three different names in three folders. A PNG photo duplicated as a fake scanner output named scan0001.png. A 1 MB file copied under a different extension, to see whether the tool trusts names or content. And inside a bulk folder of 1,000 random files, three more hidden pairs.

The decoys: a copy of the text report with exactly one byte changed, and two files sharing a name but with completely different content. A good scanner must ignore both.

Duplicate files: a clean sweep

Czkawka found every planted group and nothing else. The triple-copy report, the renamed photo, the pair with mismatched extensions, and the three pairs buried in the bulk folder. It compares content hashes (BLAKE3 by default), so names and extensions mean nothing to it. The one-byte edit was correctly treated as a different file, and the same-name impostors were ignored. That is exactly the behavior you want before trusting any delete button.

Two things to know before running it on your own disk. First, the default minimum file size is 8 KB. Our 314-byte duplicate config file stayed invisible until we lowered the threshold. That default is sensible for big cleanups, but it will hide small duplicates. Second, speed: the full scan of 1,040 files took about a tenth of a second. Our set is tiny and synthetic, so read this as a sanity check, not a disk benchmark, but the engine is clearly fast.

Similar images: good, with one caveat

Next we planted a photo along with three derived versions: a brightened copy, a half-size resize, and a JPEG re-save. At default sensitivity, Czkawka grouped all four. It also paired a graphic with its watermarked variant, and stayed silent on eight unrelated distractor images. A separate run found the three empty files we had scattered around.

Push the similarity setting to its strictest level and the resized and recompressed copies drop out. So the default is the right starting point: loose enough to catch edited versions, tight enough to avoid noise. One honest warning from an early run: when we fed it eight graphics that all shared the same dark-background-with-shapes style, it lumped them together. The tool reads composition, not identity. Eyeball the results before deleting anything. Always.

Out of the lab: a real Downloads folder

A synthetic trap proves the engine works. A real folder proves it is useful, so I also ran Czkawka on my own Downloads. It surfaced duplicate photos saved in different qualities, plus a layer of obscure orphaned thumbnails that software had created years ago and never cleaned up. The results helped me sort, and they also made me hesitate over what to keep, which is exactly why the deletion step stays a human decision.

What we did not test

Krokiet's interface: we tested the command-line edition, not the app. The screenshot below is the official one from the project, shown as illustration, not as proof of our test. We also left the deletion features untouched on the synthetic set: that test was read-only, and your first run should be too.

Official screenshot from the Czkawka project, shown as illustration.

Verdict

If your Downloads folder is a graveyard of "file (1)" copies, Czkawka passed every check we could build in an afternoon, and its first run on a real folder immediately earned its keep. Verify the groups before you delete, lower the size floor if you are hunting small files, and you get one of the rare free tools that does exactly what its README claims. If you first want to see where your disk space went, WizTree maps it in seconds; then let Czkawka handle the duplicates.