Free · no account · nothing leaves your device
Remove a photo background without uploading it
Product shots on white, studio black, a green screen — drop a photo and the background is cut from the frame. The mask runs in this tab, there is no model to download, and the file never leaves your device.
Drop images here to cut their background
JPG, PNG, WebP and iPhone HEIC. Everything is decoded and masked in this tab.
Background removed
Slide right when colour that looks like the background survives in patches around your subject; slide left when part of the subject starts going with it. Auto takes the first step that clears at least a tenth of the frame.
It is a flood fill, not a neural network — and that is why it runs here
Most background removers hand your photo to a machine-learning model. The model is good: it learned what a person's hair edge looks like. It is also several megabytes of weights, it usually wants the image on a server somewhere, and on a phone that is a slow download before the first result appears. This tool does something older and dumber. It looks at the four edges of the photo, guesses the background colour from them, and then floods inward from the border through every pixel that matches — a connected-region fill, the same primitive every paint program has had since the eighties. Whatever the flood cannot reach is the subject.
That is exact when the background is flat, which covers a surprising share of the work people bring it: catalogue product shots on white, jerseys on a black studio card, video stills cut on a green screen, sky against a silhouette. It is honest about the rest: hair, fur, smoke, glass, and a background that is a photograph of a photograph will keep part of itself behind the subject, because those pixels really do match the colour they came from. Every row tells you what it did — the colour it decided the background was, how far it cut, and when it found almost nothing at all.
What the cut actually produces
- A real transparency channel. The output is PNG or WebP, so the subject still sits on a canvas and the transparency travels with it into design tools, mockups and listings. JPEG has no alpha channel at all, so it is not offered here — saving a cut photo as JPEG throws the cut away and replaces it with flat white or black.
- A soft edge, not a staircase. The hard cut leaves a stair-step outline at any size, so the mask is smoothed with two 3×3 passes before it is written. Hair sitting on a background it matched is kept as a partially transparent pixel rather than chopped in half.
- Spill pulled back. On a green or cyan screen the subject's outline keeps a rim of the screen colour. The tool measures how much of that rim is there and pulls it back in proportion to how much of the pixel survived the cut.
- Paths through the subject survive. The fill only reaches the background from the border, so a gap between two arms, a hole in a guitar string, or the loop of a hanging cord stays opaque.
If the cut looks wrong, the slider is the knob
Auto is not a lie, it is a starting point. It takes the lowest step that clears at least a tenth of the frame, which is right for a clean product shot on white and wrong for a photo whose background falls away in a gradient, or whose subject sits close to the colour of the wall behind it. Two symptoms are worth naming, because both have a direction. Background surviving in patches around the subject means the tolerance was too tight for the light in the room — slide right. Part of the subject going transparent along with it means the tolerance was too loose — slide left. The number the tool used is printed on every row, so a cut that worked can be reproduced instead of re-guessed from scratch.
How it behaves on a phone
Large photos are masked on a downscaled copy of themselves and the mask is then scaled back up — the picture you download is still the full frame, but a phone never has to hold a flood-fill stack for twelve million pixels. HEIC files are decoded by the same decoder this site uses for its other tools, served from this origin rather than from a third party, so an iPhone photo can be cut without being converted first. Results land in your downloads one by one; on desktop and Android a batch can also be taken as a single ZIP, and iPhones get per-file downloads instead because iOS does not unpack ZIPs into the photo library.
Related tools
Once the background is gone, name the set with the Batch renamer; rewrite what is left inside the file with the Metadata editor; shrink the result with the main compressor or Compress for email; switch the iPhone format with HEIC to JPG; check what the original still reveals with the EXIF viewer; and strip the location data afterwards with Remove GPS from a photo.
Frequently asked questions
How does it decide which pixels are the background?
It samples the edge of the photo to guess the background colour, then floods inward from all four borders through every pixel that matches that colour. Whatever the flood cannot reach is the subject and is kept. A neural network would draw a finer hairline than this does; the trade is that nothing has to be downloaded, and the whole mask engine is one 9 KB file that runs on your own device.
What does the cut strength slider change?
Tolerance is how far a pixel may sit from the background colour before it counts as background. The engine measures it as the straight-line distance between the three channel values, so a step of 16 accepts only near-identical pixels and a step of 120 accepts anything reasonably close. Auto starts at the smallest step and goes up until the cut clears at least a tenth of the frame. Background surviving in patches around your subject means the step was too tight for the lighting; part of the subject going with it means the step was too loose. The exact number the tool used is printed on every row.
Is this AI background removal, and what does it fail at?
No — there is no model involved. That means it is exact on flat backgrounds (product shots on white or grey, studio black, green screen) and honest about failure elsewhere: hair, fur, smoke, glass and a busy photographic background keep part of the background behind them, because those pixels genuinely match the colour they came from. The tool reports how much of the frame it cleared, so you can see immediately whether it worked rather than discovering it after the download.
Are my photos uploaded to a server?
No. There is no upload endpoint on this site at all. The photo is decoded in this tab, the mask is computed in this tab, and the result is written to a file on your device. iPhone HEIC files are decoded by a decoder served from this same site, so even those never leave the phone.
What format do I get, and does it keep transparency?
PNG or WebP, both of which store a real per-pixel alpha channel, so the cut follows the subject's own edge and the subject is still on the canvas. JPEG cannot store transparency at all, which is why it is not offered here: a transparent PNG saved as JPEG turns the background into black or white and the cut is thrown away.
Why is the result smaller or bigger than the original?
Bigger, usually: the subject occupies fewer pixels than the whole frame did, but PNG stores that shape as a transparency mask, which costs more than the flat background did. Very large photos — above about four megapixels — are masked on a downscaled copy and the mask is then scaled back up, which keeps a phone from running out of memory; the picture itself is still written at full size.
Does it work with iPhone HEIC files and on a phone?
Yes to both. HEIC is decoded locally by the same decoder this site uses elsewhere, so the transparency you get is the same you would get from the JPEG. On a phone each result has its own download button; on desktop, and on Android, a batch of results can also be taken as one ZIP archive. iPhones get per-file downloads instead, because iOS does not unpack ZIPs into the photo library.