Change BG photo online Any background, one tap away.

A photograph of a person contains two things that have nothing to do with each other: the person, and whatever happened to be behind them. This page separates the two and gives you the second one back as a choice. The subject is left as photographed unless you ask otherwise — it can be moved and resized on a new canvas, and its brightness and warmth nudged toward the new light — and the ground it stands against becomes a flat colour you pick, a studio gradient, a photograph of your own, or a softened version of the room that was there.

The reason most people arrive is a form. Indian passport photographs are specified against plain white; state commission and university applications repeat the requirement; bank recruitment notices ask for a light background and usually name white. None of those bodies cares how you got there, only that the pixels behind the face are the value they asked for. That is a narrow, well-defined job, and this page does it rather than doing eleven adjacent ones badly.

The page opens on the way to the outline that costs nothing: colour and edges, worked out by arithmetic in a worker thread on your own machine. The Subject panel offers three Apache-2.0 licensed models beside it — a 6,163,938-byte portrait segmenter that runs in about 90 milliseconds, a 25,888,640-byte portrait matting model for hair and veils, and an 88,117,930-byte any-subject model for products and pets — each offered with its exact size on the button that fetches it, checked against its published SHA-256, cached afterwards so a second visit fetches nothing, and deletable from the same panel. Whichever you use, the outline feeds the same feather, the same edge unmixing and the same brushes, and the photograph itself never leaves this tab.

The working surface is laid out as the scene it builds: a stack of layers on the left — subject, its edge, its light, the shadow, the background and the canvas — the picture in the middle, and the settings for whichever layer is selected on the right. Split puts what you opened and what you are about to save in the same frame under a divider you drag; holding the space bar shows the original whole.

1. 1Open the photographDrag it in, click the frame, or paste it — a screenshot from a video call is a perfectly ordinary input here. Orientation is settled from the file's own flag before anything else happens, so a photo taken sideways is upright before the selection is worked out rather than after.
2. 2Look at the selection before the groundSwitch the view to Cutout to see the subject alone over a checkerboard, or press K or E to see the selection painted over the photograph in red. Red is what will be replaced. Most portraits against a plain wall need nothing at this stage; a busy room usually needs two or three strokes, or one of the models in the Subject panel, each of which states its download size before it fetches anything.
3. 3Correct it with the brushesKeep paints back into the subject, erase takes something out of it. The square brackets change the brush size and shift plus a bracket changes how soft its edge is. Cmd-Z steps back through strokes and settings alike, and Shift-Cmd-Z puts them back.
4. 4Choose the groundThe swatches are named for the job — document white for a passport or an exam form, portal off-white for the upload checks that reject a perfectly even white, listing mid grey for a product. Anything else goes in the value field. Studio grounds, a photograph of your own and a blur of the original room sit beside them.
5. 5Re-ground and saveMatch the subject's light to the new ground, and add a contact shadow if the subject is standing on something; leave it off if they are not. Then pick the canvas — the photograph's own size or a listing, profile, feed, story or banner frame — and save as JPG, PNG or WebP. The file is written at the canvas' real dimensions, not at the size of the preview you were looking at.

JPG, PNG, WebP, GIF, BMP and ICO are decoded by the browser itself; TIFF opens in Safari, which decodes it, and not in other browsers yet. HEIC — what a recent iPhone writes unless it has been told otherwise — needs a WebAssembly decoder that is fetched the first time one of them is opened, so nobody who never opens a HEIC pays for it. The models and the runtime that executes them are fetched on the same terms and no others: on request, with the figure shown first, and never otherwise. An animated GIF or WebP contributes its first frame. A file that already carries a real alpha channel is taken at its word: the selection it states is used, and the growth pass is skipped entirely.

Above roughly 32 megapixels — 16 on an iPhone or iPad, where the canvas ceiling is far lower and failing silently — you are offered a lighter working copy and told what size it will be. That is a question with two reasonable answers, not an error. Above 100 megapixels — 40 on an iPhone or iPad — the page declines and gives you the figure. Files stop at 120 MB.

Output is JPG or WebP with a quality control, or PNG. The readout on the Save button carries the format and the dimensions the file will have, and gains the byte count once you have saved one. With a kilobyte window switched on, the encoder searches quality until the file lands inside it, and the receipt reads the result back. With No ground switched on, PNG and WebP are written with an alpha channel instead of a ground. No camera or location metadata is carried across; a density in DPI is written only when you or the rule ask for one.

Routes open this same surface with the commonest jobs already set up: one on the value a form means when it says white, one on the ground each authority names for a document photograph, with the wording it used printed beside it, one on the full colour picker, one on the ID-form rules with their pixel sizes and kilobyte windows, one on the twelve studio grounds for listings and profiles, and others for transparent cutouts, product listings, profile pictures, signatures, green screens, shadows, a backdrop library and whole folders at once.

The written pages go after what a control panel cannot explain — why a photograph taken against a white wall is refused by a checker that measures pixels, the four things that give a replaced ground away to somebody who cannot say what is wrong with it, and how much of an outline a photograph actually contains before a brush has to finish the job. They are indexed on one page.

• White backgroundOpens on document white, with the off-white escape hatch beside it.
• Passport photo backgroundWhat each authority asks for, and the ground that satisfies it.
• ID photo backgroundPick the form, and the ground, pixel size and kilobyte window follow — with a crop frame and a check on the saved file.
• Background colour changerAny colour behind the subject: a full picker, hex, RGB and OKLCH, and an eyedropper that reads your own photo.
• Studio backgroundsTwelve seamless studio grounds, previewed with your own subject and matched for light and shadow.
• Transparent backgroundThe subject saved with nothing behind it: PNG or WebP with real alpha, edge cleaned first.
• Batch background changerOne recipe across a folder, every file measured, a ZIP with the receipts.
• GuidesThree written pages: the form rules, the giveaways, and what an outline can hold.
• White background rules for Indian formsWhat plain white means across passport, commission and bank portals.
• Why a swapped background looks fakeFour giveaways, each with the control that fixes it.
• Hair, fur and glassesWhat an edge can and cannot be recovered from.

The image you open is read by this page and by nothing else — there is no server behind this domain that could receive it, the outline is worked out inside your own browser either by arithmetic or by an on-device model, and if you ask for a model it is the model that is downloaded to you rather than your photo being sent to it.