日本語English

Duplicate Similar Images

Load a photo folder and identical images and near-identical photos are grouped, with one image per group recommended to keep and the reason shown. Nothing is uploaded, and your original files are never deleted or changed.

On-device・No upload

Your files and input are processed only inside your browser on this device and are never uploaded to our servers.

Drop a photo folder or images to find duplicates and look-alikes

or

Images are not uploaded; they are compared on this deviceAccepts: Images (whole folders OK)Paste (Ctrl+V) also works

Limits: 100,000 files / 100 GB to load, 20,000 images analysed (images are recognised by content)

To tidy a PC’s folders by exact matches of any file type, not only photos, use Duplicate Files

distance ≤ 10

Lower is stricter; 0 means only images that look identical even after resizing. You can change it after the search.

Ignore patterns

One per line; * and ? work. A trailing / matches folders only, a pattern with / matches part of a path, and lines starting with # are comments. After removing a pattern, load the files again.

How to use

  1. Drop a folder of photos or press “Choose folder”. You can also pick several images or add them in rounds. “Try a sample” shows how it works.

  2. Press “Find duplicates and look-alikes”. Images are recognised by content, identical ones are found, then look-alikes are analysed, with progress and a Cancel button.

  3. Use the “Similarity” slider to set how alike photos must be. Groups update immediately, without searching again.

  4. Choose the “Keep criterion” (sharpness, resolution, file size or capture date). Each group shows why its image was recommended.

  5. Compare the images of a group and pick another one with the radio buttons if you disagree. Select two images and open “Compare” to overlay them with a slider.

  6. Save the keep/duplicate verdicts as CSV or JSON, or save only the images to keep as a ZIP, as the original files. Delete in your file manager.

What it does

A keeper recommended by four criteria

The criterion you pick (sharpness, resolution, file size or EXIF capture date) decides first; ties go to the remaining criteria, then to the older file and the path order. Sharpness differences within 10 % count as the same level, so a re-save cannot reorder two equally sharp pictures.

The reason is shown

For example “Sharpest (sharpness 182) · Largest resolution 4032×3024”: only the aspects in which the chosen image is the best and better than some other image are listed. Aspects with no difference are left out. A group you changed by hand says “Chosen manually”.

Sharpness is the variance of the Laplacian

Each image is scaled to 256 px on its longest edge, converted to luma, and the variance of its 4-neighbour Laplacian is computed. Out-of-focus or shaky images score low. Scaling to a common edge lets images of different resolutions share one scale. It runs in a Web Worker, so the page stays responsive.

Identical and look-alike images in one list

Byte-identical files are confirmed with BLAKE3 and look-alikes are found with two dHashes (horizontal and vertical). Exact copies join their representative’s group, so the same photo never shows up as a separate row. The threshold (0–20 differing bits) can be changed after the search.

Side-by-side and overlay comparison

The images of a group are listed with thumbnail, resolution, size, sharpness and capture date. Pick two to overlay them and drag the divider. EXIF orientation is applied.

The ZIP keeps the original files

The images to keep are zipped as the original files, never re-encoded, so quality is unchanged. You can choose whether images that are in no group are included. A browser cannot delete your files, so use the CSV verdicts to delete in your file manager.

Use cases

  • Keep only the unblurred frame of a phone burst
  • Tidy copies that piled up through resizing, re-saving or messaging apps
  • Choose the best shot from many takes of the same scene at a trip or event
  • Find identical images hidden in a backup and review a keep/remove list as CSV
  • Tell low-resolution or re-compressed copies apart in a folder of image assets

Supported formats

Input

JPEGPNGWebPGIF (first frame)BMPAVIF and HEIC (if your browser can open them)Dropping or choosing folders, choosing several files

Output

CSV (UTF-8 with BOM)JSONZIP of the images to keep (original files)

Capability matrix

FeatureSupportNotes
Finding identical imagesSupportedSize → first/last bytes → full BLAKE3
Finding resized, re-saved or lightly colour-adjusted photosSupportedTwo dHashes (horizontal and vertical); threshold adjustable from 0 to 20
Recommending a keeper (sharpness, resolution, size, capture date)SupportedSwitchable criterion, with the reason shown
Semantic similarity (the same subject from another angle or composition)Not supportedOnly images whose pixel brightness patterns are alike are found. No AI judgement is made
Finding cropped, strongly rotated or composited imagesPartialIf the look changes a lot, the image counts as a different one
Loading HEICPartialOnly if your browser can open it (mostly Safari). If not, identical files are still found and listed, but without sharpness, resolution, similarity or thumbnails
Saving verdicts (CSV, JSON) and a ZIP of the images to keepSupportedThe ZIP holds the original files, up to 500 MB in total
Deleting or moving your original filesNot supportedA browser cannot delete files safely, so only the verdicts are exported
Videos, RAW files and other non-imagesNot supportedOnly formats the browser opens as images are covered
Offline useSupportedWorks without a connection once loaded

Limits & notes

  • Up to 100,000 files and 100 GB in total can be loaded; anything beyond is not loaded and the screen says so.
  • At most 20,000 images (in path order) are analysed; the rest are left out and counted on screen.
  • The ZIP can hold up to 500 MB in total. Above that, work from the CSV verdicts instead.
  • Nearly single-colour images, images with a brightness change in one direction only, and plain backgrounds with a few small shapes cannot be told apart by a dHash; they are left out of the look-alike check and counted on screen.
  • Sharpness is an estimate from a scaled-down image. A photo whose subject alone is in focus can score lower than another image with a busy background. Make the final call by looking at them side by side.
  • The capture date exists only if the image has EXIF data. Without it, the “Capture date” criterion compares modified times.

Privacy & safety

Images are never uploaded: your browser reads and compares them directly on your device. File names, paths, contents and hashes are not sent, stored or included in analytics. Thumbnails, the CSV, JSON and ZIP are all created on your device.

FAQ

How is this different from Duplicate Files?
Duplicate Files works on every kind of file and is meant for tidying a PC’s folders by finding files with identical content. This tool is for photo clean-up: it groups burst shots and resized or re-saved copies that look nearly the same, recommends one to keep, and lets you compare them.
How is the image to keep chosen?
Your chosen criterion comes first; ties go to sharpness, resolution, file size and capture date in that order, and then to the older file. Sharpness is the variance of the Laplacian of a scaled-down image; a higher value means crisper. The reason is shown for each group and you can override it.
Does it find the same subject photographed from another angle?
No. Only photos whose brightness patterns are alike are grouped. The same person or place with a different angle, composition or pose counts as a different photo. No AI judgement of semantic similarity is made.
Can this tool delete the duplicate images?
No. A browser has no safe way to delete your files. Save the verdicts as CSV or JSON, or save only the images to keep as a ZIP, and tidy the original folder in your file manager. The ZIP contains the original files, not re-encoded copies.
Can it use HEIC photos?
If your browser can open HEIC (mostly Safari), they are analysed like any other image. If not, identical HEIC files are still found and listed, but without sharpness, resolution, similarity or thumbnails.
How many images can it check?
Up to 100,000 files and 100 GB can be loaded, and at most 20,000 images are analysed. Files are read one by one so memory does not grow while analysing, but many large photos take time. When saving a ZIP, files are read and written one at a time. Progress is shown and you can cancel.

Related tools

Duplicate Files

Load a whole folder and see groups of files whose content is byte-for-byte identical, with the space they waste. Nothing is uploaded and nothing is deleted: save the keep/duplicate verdicts as CSV or JSON and clean up with your own file manager.

Image Compressor

Drop images to shrink them in bulk while keeping them sharp. Convert to WebP or AVIF, resize, and compare before and after. Your images are never uploaded.

Batch Image Resizer

Drop a batch of images and resize them by width, longest edge, box crop and more. Make several sizes (say 320, 640 and 1280) from each image in one go, then copy ready-made srcset HTML. Your images are never uploaded.

EXIF & GPS Remover

Remove the location (GPS), camera details and timestamps embedded in your photos without touching image quality. Inspect what was inside and verify the result — all in your browser.

Icon Generator

From one image (or a letter or emoji), get favicon.ico, apple-touch-icon, Android/PWA icons, a maskable icon and an OGP image, plus the manifest and the <head> code to paste. Check blur at real size first — nothing leaves your browser.

Image Converter

Choose the target format first, then drop images to convert them in bulk. Tune quality per format, use lossless WebP, set a background for transparency and name the output files — nothing is uploaded.

Last updated: