Duplicate Image Finder
Find duplicate images in your collection
Click or drop several images hereChoose several images
About Duplicate Image Finder
Duplicate Image Finder compares a set of pictures you select and lists the files that are the same, one line per group of copies. Identical files are found by their SHA-256 fingerprint. The same picture saved differently, resized, re-compressed, or converted from PNG to JPG, is found with a difference hash (dHash, the method of the imagehash library): each image is shrunk to a tiny gray thumbnail and turned into 64 points (bits), and two images count as the same picture when only a few points differ. The hash is also taken in the 8 orientations of the picture (as it is or turned by 90°, 180°, or 270°, each mirrored or not), so a rotated or mirrored copy is found too. The tool only lists the copies; it never deletes or moves your files. The fingerprints and hashes are computed in your browser.
Example
Eight files, one of them a different photo: the page groups the copies, the rotated and mirrored ones included, and leaves out the other photo. Measured with this page's comparison code on a 1335 × 2000 photo; a browser shrinks pictures its own way, so it can count a point or two differently.
Files selected
IMG_2041.jpg the original photo IMG_2041 (1).jpg a byte-for-byte copy IMG_2041-half.jpg resized to 50% IMG_2041-q30.jpg re-saved at JPG quality 30 IMG_2041.png converted to PNG IMG_2041-rot.jpg rotated by 90° IMG_2041-mirror.jpg mirrored left to right IMG_2187.jpg another photo
Result at the Normal level
Same picture (each picture matches another one in this group on at least 63 of the 64 points compared): IMG_2041.jpg = IMG_2041 (1).jpg = IMG_2041-half.jpg = IMG_2041-q30.jpg = IMG_2041.png = IMG_2041-rot.jpg = IMG_2041-mirror.jpg Identical files: IMG_2041.jpg = IMG_2041 (1).jpg Not listed: IMG_2187.jpg
How to find duplicate images
- 1Click the upload box and select several images at once; a new selection replaces the previous one.
- 2Pick a level in "Same picture when": Strict, Normal (the default), or Loose.
- 3Click "Find duplicates".
- 4Read the list, one line per group of copies: red "Identical files" lines are byte-for-byte copies, amber "Same picture" lines show how many of the 64 points compared match; delete the extras yourself.
Formats and limits
- Input
- Two or more pictures the browser can open (JPG, PNG, WebP, GIF, BMP, AVIF); others are compared for exact copies only
- Levels
- Strict: at most 3 of the 64 points differ; Normal: at most 6; Loose: at most 10. Rotated or mirrored copies: at most 3, 6, and 6
- Not found
- At the Normal level: copies turned by a small angle, such as a straightened photo (2 of 18 found), and most copies cropped by 5% or more on each side
- Very large pictures
- Above 268 MP a picture is not decoded and is compared for exact copies only
- Empty files
- Left out of the comparison, with a message naming them
Common uses
- A camera roll imported twice
- After importing the same phone photos to a computer twice, select the folder's pictures: copies such as IMG_1234.JPG and IMG_1234 (1).JPG show up as red "Identical files" lines, whatever their names.
- Photos that came back through a chat app
- A photo sent through a messaging app usually returns resized and re-compressed, so its bytes no longer match the original. At the Normal level, the pair is listed as "Same picture" with its count of matching bits.
- One product shot saved as PNG and as JPG
- A store's image folder often holds the same product shot in two formats and two sizes. The perceptual hash ignores both, so such pairs are listed as "Same picture" and you can keep the format your site uses.
- Repeated screenshots
- Two screenshots of the same screen are rarely identical byte for byte, since a clock or the cursor can differ. Their hashes still match closely, so the Strict level catches them while listing fewer look-alike pictures.
Which tool to use
- Compare Images: takes exactly two pictures and shows where they differ, in a slider view or with every changed pixel painted red and counted; pick it to inspect a "Same picture" pair, not to search a batch.
- Compare Files: compares any two files of any type byte for byte and gives the position of the first difference; it never decodes a picture, so a resized copy counts as different.
- Hash Generator: computes SHA-256, the fingerprint this page uses for identical files, and 16 other hashes of any file; pick it to check one download against a published checksum.
Where your image is processed
The SHA-256 fingerprints and the perceptual hashes are computed by this page in your browser; the pictures are not uploaded and nothing on your device is deleted or moved. The list vanishes when you close the tab. If a message is displayed, its text, stripped of file names, is reported to us along with the tool's name and the browser's name and version.
Frequently asked questions
Does it find resized or re-saved copies?
Yes. Besides byte-identical files, it compares a perceptual hash of each picture, so the same photo resized, saved at another JPG quality, converted to another format, rotated by 90°, 180°, or 270°, or mirrored is listed as "Same picture" with its count of matching points.
Which copies does it actually find?
We ran the page's comparison code on 6 photos, 1297 to 2000 pixels wide, and 21 kinds of copies, shrinking each picture with three different resizing methods, since browsers do not all shrink alike: 18 runs per kind of copy. At the Normal level it found every copy that was byte-identical, resized to 50% or to 320 pixels wide, resized to 1280 pixels and re-saved at JPG quality 70 as a chat app does, re-saved at JPG quality 30, converted to PNG, turned to grayscale, given a small text watermark, rotated by 90°, 180°, or 270°, mirrored left to right or top to bottom, or rotated by 90°, halved, and re-saved at quality 70 (18 of 18). It found 17 of 18 copies re-saved at JPG quality 60, brightened by 10%, or cropped by 2% on each side; 16 of 18 converted to WebP at quality 75; 8 of 18 cropped by 5% on each side; 1 of 18 cropped by 10%; 2 of 18 rotated by 3°. Against 886 pairs of different pictures (photos, quarters of photos, cut-out objects, and generated test pictures, 2,658 comparisons), it made no false match at the Normal level; the closest pair was 10 points apart.
Which level should I choose?
6 differing points is the Normal level, the default, and it suits resized or re-compressed copies. Strict accepts three and lists only near-identical pictures; Loose accepts ten and also catches lightly edited copies, with more false matches. In our test, Loose found 16 of 18 copies cropped by 5% on each side against 8 of 18 at Normal, and matched 1 of 2,658 comparisons of different pictures (two quarters of different photos, 10 points apart), which Normal and Strict never did; Strict missed 4 of the 18 copies shrunk to 320 pixels wide.
Can it be fooled?
Yes, both ways. At the Normal level it misses most copies cropped by 5% or more on each side, and a photo turned by a small angle (rotated by 3°: 2 of 18 found), since only quarter turns and mirror images are tried. And two very plain images can look alike to the hash: a blank page and the same page with one line of text differ on only 4 of the 64 points, so the Normal level lists them as the same picture. Likewise, the left and right halves of a symmetrical object are mirror images of each other, so they can be listed as the same picture. Transparent areas count as white. Look at the thumbnails before deleting anything.
Does it delete the duplicates?
No. It lists the copies by file name only. Delete or move the extra copies yourself in your file manager or photo app; the page cannot touch the files on your device.
How many pictures can I compare at once?
The page sets no limit. Every picture is compared with every other one, so 100 pictures make 4,950 pairs, and each file is read whole and decoded once, one after another: a large batch of big photos takes longer and uses more memory.
Can it scan a whole folder or my phone's gallery?
Not by itself: there is no folder option, and the page cannot browse your device. Open the folder in the file picker and select every picture in it; pictures kept in different folders must first be copied into one, since a new selection replaces the previous one.
Are three copies of the same photo shown as one group?
Yes. Three copies A, B, and C give one line: A = B = C. A file joins a group when it matches at least one file already in it, so at the Loose level a chain of slightly different pictures can end up in one group; look at the thumbnails, keep one file, and delete the others.
Does it work with HEIC or camera RAW files?
Only for exact copies when the browser cannot open them: Chrome, Edge, and Firefox do not open HEIC, and no browser opens camera RAW. Such files are still compared byte for byte, and the page names them.
Do I need to install anything? Does it work on a Mac or a phone?
Nothing to install and no account: the page runs in a current browser on Windows, macOS, Linux, Android, or iPhone. The trade-off against a desktop duplicate finder: it cannot scan your drives by itself or delete files, so you select the pictures and remove the extra copies yourself.
Tips and tricks
- ✓The page shows names, not file sizes: compare the sizes in your file manager to keep the larger copy.
- ✓Open a "Same picture" pair in Compare Images to see which pixels differ.