Background remover for plain backdrops — free, and the photo never leaves your device

No, and that is worth being blunt about because almost every other background remover does. The paid tools run a trained segmentation model that has learned what a person, a car or a chair looks like, which is how they cut a hiker out of a mountain range. The smallest useful model of that kind is several megabytes of weights that would have to be downloaded before you could do anything — this entire site is built to stay under about a hundred kilobytes. So this tool does something much simpler and much more predictable instead: it decides what is background by colour and by connection to the edge of the frame. On a plain backdrop the result is often indistinguishable from the expensive version, and it is instant. On a busy background it will fail, and it will fail visibly rather than quietly.

Because deleting every pixel of a colour destroys the subject. Photograph someone in a white shirt against a white wall and a global "remove white" takes the wall and the shirt together, leaving a floating head. Filling inward from the border only removes pixels that are both the right colour and reachable from the edge of the picture without crossing the subject — so the wall goes, and the shirt, being surrounded by a face, arms and hair, stays. The same logic protects a white logo in the middle of a product, the sky visible through a handle, and the white of an eye. The trade-off is that a background area completely enclosed by the subject — the gap inside a mug handle, the space between an arm and a body — is not connected to the edge, so it is left alone. Clicking inside that gap seeds the fill there too and clears it.

Well: a studio backdrop, a painted wall, a photo booth, a product on a white sweep, a scanned signature on paper, a logo exported onto a flat colour, and anything shot against a green or blue screen. Badly: a person outdoors, a room with furniture behind them, anything with a background that has texture, pattern, or a wide range of colours. In between and often fixable: a backdrop with a soft shadow or a light gradient (raise the tolerance, or click to add a second reference colour), and a subject whose colour is close to the backdrop — a grey cat on a grey wall is hard for a colour-based tool for exactly the reason it is hard for your eye. Fine detail is the other limit: wispy hair, fur, lace, netting and motion blur have hundreds of pixels that are genuinely half background, and no threshold gets all of them right.

Because the edge pixels really are part background. A camera does not record a sharp boundary — a pixel sitting on the outline of a shoulder captured some shoulder and some backdrop, and the value stored is the mixture. That is anti-aliasing, and it is what makes photographs look smooth instead of jagged. When the backdrop is removed, those mixed pixels are left holding some of its colour, which reads as a pale halo on a white backdrop or a green rim on a green screen. Tolerance trades one artefact for the other: raise it and the fringe thins but the fill starts biting into the true edge of the subject, making it look eroded; lower it and the subject stays intact but keeps its halo. The edge softness control helps by ramping transparency across that mixed band rather than cutting through it, which is why partly-transparent pixels exist in the output at all. If the fringe still shows, putting a solid colour behind the subject hides it far better than transparency does, and choosing a background colour close to the original backdrop hides it almost completely.

Because green is the colour furthest from human skin, and distance in colour is the entire basis of this kind of cut. Skin tones, hair, wood, most clothing and most product packaging sit in the red-orange-brown part of the spectrum; a saturated green sits a long way from all of it, so the threshold separating background from subject has a wide, forgiving gap to sit in. Broadcast chroma green also happens to be the channel digital sensors record with the most detail, which means the edge is defined more precisely there than it would be on blue. White is a much harder backdrop despite looking easy, because white is close to every pale thing in the frame — a white shirt, blonde hair, a highlight on a cheek, the shine on a plastic bottle. If you are choosing a backdrop and the subject is not green, green wins. Blue is the fallback for green subjects and for anything with green in its branding.

Yes, by a wide margin, and it is worth saying even though it is not what a background remover page is supposed to say. Every problem on this page — fringing, eroded edges, hair that will not cut, a subject the same colour as its backdrop — is created at the moment the shutter fires and can only be reduced afterwards, never undone. Three things do most of the work. Put distance between the subject and the backdrop, so the backdrop falls out of focus and, more importantly, so the subject does not cast a shadow onto it; a shadow is a second background colour and the most common reason a cut looks patchy. Light the backdrop separately from the subject, so it reads as one even tone rather than bright in the middle and dark at the edges. And avoid dressing the subject in the backdrop colour. Ten minutes of setup beats an hour of tolerance-nudging, and it is the difference between a cut that is clean and a cut that is merely acceptable.

When the background is not plain. If there is furniture, a street, foliage, a crowd or a landscape behind your subject, no threshold on this page will save it, and pushing the tolerance up until the background disappears will take half the subject with it. The same goes for close-up portraits where the hair is the point, and for animals — fur is thousands of half-transparent pixels and needs a model that knows what fur is. Commercial tools trained on segmentation handle all of that, and for one important image the cost of a single credit is less than the time you would spend fighting a slider. Use this page for what it is good at, which is the large and boring category of photos already shot against something plain — product shots, headshots on a backdrop, logos, scans, ID photos, listing images — and it will do those in a second without an upload, an account or a watermark.

Nothing is uploaded and nothing is saved. The file is read into this tab, decoded into pixels, processed by JavaScript running on your own device, and offered back to you as a download. It never touches a server, so it also works with the connection switched off after the page has loaded, and closing the tab is all it takes to be rid of it. This page writes nothing to local storage either — not the picture, not your settings — so there is no copy left behind on a shared computer. That matters more here than on most tools: people put photographs of themselves, their children and their identity documents into background removers, and the usual model is to upload the image to somebody else’s server and trust a policy about what happens to it next. The one thing the page does load from outside is the advertising, which is the same on every page of this site and is described on the privacy page.

Comfortably up to about thirty megapixels, which is larger than most phone and camera output. The limit is memory rather than speed: a picture held for editing is four bytes per pixel, so a thirty megapixel photo is a hundred and twenty megabytes for the source alone, before the mask, the fill queue and the finished copy. Above that a phone browser gives up. There is a second, lower ceiling that has nothing to do with this page: Safari on iOS silently returns a blank canvas above roughly sixteen and a half million pixels — no error, no warning, just a white image — so pictures above that are scaled down before processing and the page tells you when it has done so. The preview is also computed on a reduced copy so the sliders stay responsive; the file you download is always processed at full resolution, which is why saving takes a moment longer than moving the slider does.

PNG because it is the common format that carries an alpha channel — the extra value per pixel that records how transparent that pixel is. JPEG has no alpha at all. Hand it an image with transparency and it does not warn you; it flattens the transparent areas to a solid colour, usually black, and you find out when you drop the picture into a slide. So while the background is transparent, PNG is the only honest option and it is the only one offered. As soon as you choose a replacement colour the picture becomes fully opaque, there is no transparency left to lose, and JPEG becomes a reasonable choice — usually a much smaller file for a photographic subject. That is exactly when the format appears in the list, and not before. If the PNG comes out larger than you expected, that is normal: PNG is lossless, and a photograph with a cut-out subject stores badly in it.