Remove the background (remove bg)
The fast version: the page opens with white preselected, so a drop and a generate give you the cutout on a clean white board — or switch to transparent PNG for anything that needs to sit on top of something else.
Right removes more background; left keeps more of the subject.
- Nothing uploaded
- Free, no sign-up
- Strips EXIF
- Batch files
There are three moments where you need someone's background gone and you need it now, not after a sign-up wall. The visa application is open and the photo must have a plain background, and the one you have is a selfie against a bookshelf. The profile picture for a new work account is the holiday photo, and the sea behind the face is the wrong answer to a form that asks for head and shoulders on white. The group chat is asking for your face to put on the team roster, and the one decent shot has a café wall behind the head. In all three, the picture is fine and the room is not, and the fix takes less than a minute.
This page is built for that minute. It opens with the white background already chosen — the one a passport frame, an ID sheet, an e-mail signature and most government photo rules actually want — so the entire job is: drop the photo, generate, take the file. If the shot needs to sit on top of something rather than stand alone, flip the control to transparent and you have the see-through PNG of the same cutout. The AI that decides what is you and what is the room runs inside your browser, on your device, after a one-time model download of about 4.5 MB.
One photo at a time, one clean result, no account, no credit, no watermark at the edge of the cutout. The photo you drop in is not sent anywhere, and the photo you download is the whole thing — no 500 px trial file that unlocks at a paywall size.
How it works
Two clicks between a backgrounded selfie and a clean cutout on white.
Drop the photo
A JPG, PNG or WebP of the person or object — the face or product should fill a good part of the frame, and the background should be plain enough that a stranger could name it. The page reads the file on your device and prepares the cutout; there is nothing to configure and nothing to upload.
Generate
Press generate. The model, already cached after your first visit, decides where the subject ends and the room begins and draws the edge. If the result is a hair off — a sliver of wall kept, a corner of a shoulder dropped — the level slider shifts the border and redraws it live without re-running anything.
White, or switch to transparent
The export lands on white by default, which is the format a passport rule, an ID sheet and a print shop each want named. For an avatar, a roster tile or anything that will sit on a colored layout, tap the transparent option before you save and the same cutout comes out as a see-through PNG instead.
The two outcomes, and what each one is for
White is the answer for every use that wants a finished picture. A passport-style frame prints from a white board, not from a see-through gap. A headshot for a company page that has its own background color usually still wants white behind the shoulders, because a transparent gap under the chin is the tell of a cutout. A print order for a badge, a certificate or a photo book wants flat, predictable, white.
- White: passport and visa frames, ID and exam sheets, work-profile headshots, printed badges
- White: any layout that already has a white area behind the photo slot
- Transparent: avatars on colored app headers, roster and team-wall tiles, design comps, cutouts for a collage
- Either: the choice changes the file, not the cutout — the same edge, two exports
What the model actually sees
The U²-Net model does not look at your photo the way you do. It looks at a 320-pixel thumbnail of it, and what comes back is a grayscale map of confidence — where it is sure the subject is, where it is unsure, where it is sure it is background. The page draws the edge along that map. The practical result: a face against a wall separates perfectly, because at thumbnail scale the answer is obvious; hair against a bright sky is where the map gets unsure, and that is where the level slider earns its place instead.
- Sure-yes and sure-no regions come back clean, with no intervention
- The unsure band — hair, fur, thin limbs, a sheer sleeve — is where the slider moves the border
- The edge is a soft mask, not a vector path: pixel-exact borders are not what this tool produces
The level slider, in one sentence each
It is a threshold sliding across the confidence map, not an eraser dragging across the picture. Slide one way and the pixels the model was unsure about get kept; slide the other way and they get dropped. Because the redraw is just the same map re-thresholded, it keeps up with your finger on a phone screen — drag until the result reads right, and the cost of each position is zero.
Two tells point the direction: a faint ring of the old background hugging the edge means the border is too generous, so slide to drop it. A nick taken out of dark hair or a black collar means the border is too hungry, so slide back to keep it. The setting is global — one value for the whole frame — so a fix in one corner moves the edge everywhere, which is why a tighter crop around the subject beats a heroic slider position.
Why there is no queue here
Each cutout runs the model on your own processor, and that is what makes one photo at a time the honest shape of the tool. There is no upload to queue, no plan tier that adds a batch slot, and no size wall after the third image. If you have a folder of faces, it is a folder of one-minute passes, in the order you feel like doing them — and the reason the per-photo shape is a feature rather than a limit is that the photo never left the room it was taken in.
Frequently asked questions
My visa form wants a white background photo. Does the cutout work for that?
Yes, and white is what the page opens on, so the default result is the one the form is asking for. Crop to the head-and-shoulders frame the form describes first, then run the cutout — the white board comes out as part of the PNG. If the form also lists a file-size maximum, run the result through one of the KB target compressors; the white stays white through a PNG or WebP pass.
Why is white preselected if transparent exists?
Because the most common answer to remove-bg searches is not a design cutout, it is a photo for a form — and those forms want a flat background, not a hole. White is the default because it is the more common second half of the job. Transparent is one tap away when the photo is going onto a layout instead of under a rule.
I took a photo in the car for my profile. Will it look like I removed the background, or will the edge betray it?
The edge quality depends on what is behind you, not on this tool: a plain grey car seat behind a dark jacket gives the model a clean boundary and the cutout reads as studio. A busy window scene gives it noise to argue with, and a soft band can show up around the hair. The slider cleans the obvious cases; the truly rough ones want a plain wall and a second take, and that advice is cheaper than a design retainer.
Can I get the same person cut out on white and on transparent?
Yes — run the cutout once, generate the white version, then flip the control to transparent and generate again. The model runs twice, the edge is the same both times, and you end up with the two files a typical profile actually needs: the one for the form and the one for the layout.
Is there a size limit on the output?
The long side caps at 2048 pixels, which is already past the resolution a passport frame, an avatar or a web header will print or display. A 12 MP source comes out at that cap with the subject sharp; if you are after a print master at the original resolution, that is a different tool's job, and no browser-based cutout on the market is doing it for free.
Does anyone see my photo?
No. The 4.5 MB model downloads once and caches in your browser; the photo itself is segmented and saved on your own machine. There is no account to see it, no server copy to delete, and no usage counter to trip — the only thing that crosses your connection in either direction is that one model file, on the first visit.