How to cut out an object
- Add a picture. Drop one anywhere on the page, paste it, or click the box to choose a file. Phone photos, including iPhone HEIC, work too.
- Wait a few seconds. The first time, the page downloads the selection model (about 45 MB, which your browser keeps), and then it studies your picture. After that the page says Ready.
- Click on the object. A blue tint shows what is selected. Click again to add more, switch to Click to remove (−) to take a part out, or draw a box around the object instead.
- Check the result. Switch the view to Result to see the finished cut-out, then adjust the edge if you like.
- Press Download the result. It is made at the full size of your picture.
Each click takes a fraction of a second once the picture has been studied. Another picture starts again, and the model stays loaded.
Selecting: clicks, removal and boxes
- Click to add (+) puts a green point on the object. One click is often enough. Add more clicks on parts that were missed. Several clicks on different objects select all of them together.
- Click to remove (−) puts a red point where the selection has gone too far. You can also use Shift + click, or the right mouse button. A remove click helps where the model has selected too much, for example the whole animal when you want only its head. Inside a plain, even shape it can have no effect, because the model sees nothing to separate there.
- Draw a box and drag around the object. A box is the most direct way to say which object you mean, and it works well for objects that touch other things.
- Undo takes the last click or box back (Ctrl + Z or Cmd + Z does the same), and Clear starts the selection again.
- Size of the selection. One click can mean different things: a small part, a bigger part or the whole thing. The model proposes three guesses, and Best guess takes the one it rates highest. Small part, Medium and Whole thing choose by size. For a click on the face of a cat in our test, they selected 17%, 45% and 48% of the picture: the head, the cat, and the whole cat including the paw at the bottom. This choice is only for a single click: with more clicks or a box, you refine with points instead.
Smoothing the edge
| Control | What it does |
|---|---|
| Grow or shrink | Moves the whole outline outward (+) or inward (−) by up to 10 pixels, measured on the picture as shown. Grow a little to catch a thin outline of the object, and shrink a little to lose a halo of background. On a full-size picture the amount is scaled up to match. |
| Soften the edge | Blurs the outline slightly so the object blends in when it is placed on something else. 0 keeps the edge as the model drew it, which is already smooth because the page works out partly-selected pixels along it. |
| Clean up specks | On by default. Removes tiny stray pieces outside the object and fills tiny holes inside it, but keeps real gaps, such as the space between a handle and a cup. In our test, this removed the stippling that appeared along a cat's thin ear tip. |
| Select everything else | Flips the selection, so the result is the picture without the object. |
What you can save
| Result | What you get |
|---|---|
| Cut-out | The object with the background see-through, as a PNG or WebP. Trim to the object crops the picture to the object's box. A JPG cannot be see-through, so it is not offered for a cut-out. |
| Mask | A black and white picture the same size as yours: white where the object is, black elsewhere, with soft edges. Useful for editing programs and for training data. |
| On a color | The object laid on a plain color you choose, saved as a normal picture. |
| Blur the rest | The object stays sharp and everything else is blurred, which gives a portrait-style background. The blur amount runs from 1 to 30. |
The file is named after your picture with the kind of result added, for example dog-cutout.png or dog-blur.jpg. Your original is never changed.
How well does it work?
We tested it against pictures where we knew the exact outline. On a 1600 × 1200 test picture with a textured background:
- One click on a red disc gave an outline that overlapped the true one by 99.4%, and every pixel inside it was the picture's own, unchanged.
- Drawing a box around a blue rounded rectangle gave an overlap of 99.5%.
- Two clicks, one on the disc and one on the rectangle, selected both at 99.5%.
On a photo of a cat, the selection from one click on its face was identical to the one that the same model produces when it is run natively on a computer. We also checked the model running in the browser against that native run, and the numbers it works from agreed to within about one part in a million.
What it cannot do well
- It is a selection tool, not a hair-level cutter. The model works on a picture 1024 pixels across, and the outline is then enlarged to your full-size picture with smooth edges. Thin fur, wisps of hair, whiskers, smoke, glass and other see-through things come out approximate, with a hard-ish edge. Cut-outs of animals and people are fine for most uses, and a close look at the edge can show it.
- The first guess can be the wrong thing, especially in busy pictures. A few clicks or a box usually fix that.
- Pictures over 16 megapixels are saved at 16 megapixels, and the page tells you when that happens.
- One picture at a time. To cut out the same kind of thing from a lot of pictures, the background remover is quicker, because it works on a whole folder without clicking.
Ideas for using it
- Stickers and profile pictures. Cut yourself, a pet or a product out and save a see-through PNG.
- Product photos. Put the object on white or a brand color.
- Portrait-style blur for a photo where the background is distracting.
- Training data. Save a mask of the object, to show a training tool where the subject is. See the guide to preparing an image dataset.
- Collages. Cut several objects out, then put them together with Combine images, or crop first with the crop tool.
- Taking something out instead. To remove an object from a photo and fill the space in, use the object remover.
How it works
The page uses MobileSAM, a small, fast version of Meta's Segment Anything model (the code is Apache-2.0 licensed, and the weights and the ONNX files we load are published under the MIT license). It has two parts. An encoder studies the picture once, which takes a few seconds, and a small decoder turns each of your clicks or boxes into candidate outlines within a fraction of a second. The outline comes back as scores for each pixel, and the page turns them into a smooth, partly-transparent edge at the full size of your picture.
Privacy
- The selection is made inside this tab, and your files on disk are never changed or deleted.
- The two model files (about 45 MB together) are downloaded the first time from Hugging Face, a model hosting site, and your browser keeps a copy.
- For what this site collects and keeps, see our privacy policy.
Frequently asked questions
How do I cut an object out of a picture?
Add the picture, wait for Ready, and click on the object. A blue tint shows the selection. Add or remove clicks to refine it, then press Download the result to get the object on a see-through background.
How is this different from the background remover?
The background remover finds the main subject by itself and works on a whole folder. This page lets you choose exactly which object you want, including a part of a picture or one of several things, with clicks or a box, one picture at a time.
Can I select more than one object?
Yes. Click once on each object and they are selected together. In our test, a click on a disc and a click on a rectangle selected both.
Why did the click select too much or too little?
A single click can mean a part, an object or a bigger group, so the model has to guess. Use Size of the selection to pick another guess, add a click on a missed part, use a remove click on something that was included, or draw a box around the object.
Why is the edge rough on fur or hair?
The model works on a picture 1024 pixels across, and thin or see-through details such as fur, hair, smoke and glass are beyond what it can outline exactly. Try Soften the edge or Grow or shrink, and expect to touch up a fine edge in an image editor if you need it perfect.
Can I save the result as a JPG?
A cut-out with a see-through background needs PNG or WebP, because JPG cannot be see-through. The mask, the object on a color, and the blurred background can be saved as a JPG.
Is the cut-out the same size as my picture?
Yes. It is made at your picture's full size, up to 16 megapixels. A bigger picture is saved at 16 megapixels, and the page says so. With Trim to the object, the file is cropped to the object's box.
Why did the first picture take longer?
The first time, the page downloads the two selection model files (about 45 MB) and then studies your picture, which takes a few seconds. Your browser keeps the model files, so after that only the few seconds of studying each new picture remain.
Is my picture sent anywhere to be cut out?
No. The selection is made in your browser tab, using model files that are downloaded once. For what the site collects and keeps, see our privacy policy.