Which tool do I need?
| You want to… | Use | Good to know |
|---|---|---|
| Remove copies that were resized or re-saved | Duplicate Image Finder | No model to download. Three levels of "how alike". |
| Remove copies that were also mirrored, or cropped a little | Group Similar Images, set to Near-copies | A model of about 85 MB. Up to 4,000 pictures at a time. |
| Sort a folder by what the pictures show | Group Similar Images | A slider from "same scene" to "same kind of thing". |
| Find the pictures of something you can describe | Search Images | A model of about 146 MB. Up to 6,000 pictures at a time. |
| Keep only pictures with faces, or drop crowd shots | Face & Expression Sorter | A small detector (about 190 KB), made for photos of real people. |
Three kinds of "alike"
"Duplicate" can mean three different things, and each needs a different method:
- The same file, or the same picture saved again. The Duplicate Image Finder shrinks each picture to a small pattern of light and dark and compares the patterns. It catches resized, re-saved and recompressed copies, which an ordinary file comparison misses. It needs no model, so it starts at once.
- The same picture, changed more. A mirrored or cropped copy has a different pattern of light and dark, so the finder rarely catches it. Group Similar Images uses a model that understands what a picture shows, and in our test caught every copy we made: we took 40 different photos and added 10 copies (four resized to half, four mirrored, two re-saved as JPG at quality 35), and all 10 were grouped with their original, with no group joining two different photos.
- The same subject, a different picture. Ten photos of the same dog are not copies, but you may want to see them together, or keep two. That is grouping, and Search Images finds them by a description instead.
Our guides explain the idea in depth: how to find duplicate photos and perceptual hashing.
A workflow for cleaning a folder
- Duplicate Image Finder first, on Resized or re-saved. It is the quickest, with no download, and removes the bulk of the copies.
- Group Similar Images on Near-copies to catch what is left: mirrored and slightly cropped copies. Keep one picture from each group.
- Face & Expression Sorter if the set should contain people: split it into pictures with and without faces, and drop crowd shots by face count.
- Search or group by subject to pull out the part you need, for example every picture of a dog, or to see that you have only three kinds of scene.
- Then tag or caption the result with the image tagger tools, so you do not label pictures you are about to throw away.
The reasons for this order are in our dataset workflow guide. Always keep the downloaded ZIP and your original folder separate until you are happy with the result: the tools never change or delete the originals on your disk, and the ZIP holds exact copies under their original names.
What each one cannot do
- The Duplicate Image Finder rarely catches mirrored copies, never catches rotated ones, and only sometimes catches heavy crops. Fix the orientation first, and look at groups by eye.
- Group Similar Images and Search Images use a general model that is not perfect: it can mix up kinds of things (in our test of 80 photos, dogs and cats were the most common mix-up), it understands English words best, and it does not count well. Read the results before you delete anything.
- The Face & Expression Sorter uses a detector built for photographs of real people. It misses small, turned-away and covered faces, and stylized faces such as anime, cartoons and paintings are detected only some of the time. Expression labels are a rough signal.
- Sizes: grouping handles up to 4,000 pictures at a time, because it compares every picture with every other, and search up to 6,000. Split a bigger archive into parts.
Where the work happens
Reading and comparing happen in your browser tab. The tools with a model download it once from Hugging Face, a public model host, and your browser keeps it. Your picture files on disk are never changed. For what this site collects and keeps, see our privacy policy.
Frequently asked questions
How do I find duplicate photos in a folder?
Drop the folder onto the Duplicate Image Finder. It groups copies, even resized or re-saved ones, and you uncheck the ones you want to remove and download the rest.
Why did the duplicate finder miss a copy?
It compares patterns of light and dark, so mirrored, rotated or heavily cropped copies look different to it. Try Group Similar Images on Near-copies for mirrored and cropped copies, and fix the orientation of rotated ones first.
What is the difference between Group Similar Images and Search Images?
Grouping sorts the whole folder into piles of pictures that look alike, without you saying what to look for. Searching finds the pictures that match a description you type, such as "a dog on a beach".
Do these tools delete my pictures?
No. They never change or delete the files on your disk. You choose which pictures to keep, and the tool saves exact copies of your choice as a ZIP.
How many pictures can I organize at once?
Grouping handles up to 4,000 pictures at a time and search up to 6,000. The Duplicate Image Finder has no such fixed limit but slows down with very large folders; its page has the details.
Can the face sorter find anime faces?
Only some of the time. It was built for photographs of real people, and drawn faces have different proportions and shading. For illustration sets, check its results by eye.
Are my pictures uploaded to be sorted?
The comparing is done in your browser tab. For what the site collects and keeps, see our privacy policy.