Skip to main content
PulsarKit

Remove background from image

Cut the subject out of any photo with AI: our server finds it in a reduced 1024 × 1024 copy, and your browser makes the cut-out from the photo itself. Download a transparent PNG, or the same cut-out on a solid colour.

Drop your files here

.jpg, .jpeg, .png, .webp · up to 30 MB

How to Remove Background

  1. 1

    Add your photo

    JPG, PNG or WebP, up to 20 files at once and 30 MB each. The file itself stays on your device: to find the subject, the page sends our server only a reduced 1024 × 1024 copy of the frame, on a computer, a phone or a tablet alike.

  2. 2

    Pick transparent or a background colour

    Transparent PNG is the default. Choose white for shop listings, or any colour of your own. Edge softness and Precise edges control how the outline is cut.

  3. 3

    Click Remove and download the PNG

    Our server finds the subject in the reduced copy and returns its outline; your browser cuts it out of the photo, drops the background and gives you a PNG at the same size as the photo you gave it.

What actually happens to your photo

A segmentation model called BEN2 looks at the picture and decides, pixel by pixel, what belongs to the main subject and what belongs behind it. The answer is a mask: a greyscale picture where white is the subject and black is everything else. The model runs on our server, not in your browser, on every device. The page shrinks the frame to a 1024 × 1024 copy, which is the size the model works at anyway, sends only that copy over an encrypted connection and gets the mask back. Everything else happens in this tab: the mask is stretched to the size of your photo and becomes the alpha channel of a PNG, or, if you asked for a colour, the recipe for blending the subject onto it. Your original file is never uploaded. The reduced copy is deleted as soon as the mask is ready, and the mask within about two hours. Nothing is downloaded into the browser beforehand, so there is no slow first run. The model stays on the server because it is heavy: inside a browser tab it needs more than 2 GB of memory, which a phone cannot give it, and on many computers it takes half a minute per photo.

Transparent PNG, white, or a colour of your own

A transparent PNG is what you want when the cut-out is going somewhere else: a slide, a poster, a product page with its own backdrop. It is also the heaviest of the three, because PNG stores every pixel exactly and an alpha channel on top. White is the format most marketplaces ask for, and it compresses far better because the background becomes one flat area. A custom colour does the same thing with the shade you pick, which suits a team page or a set of avatars that all have to match. All three come back as PNG, deliberately: the format of the result should not change when you move a switch, or the next tool in the pipeline would suddenly see a different kind of file.

Edges, hair, and the Precise setting

The hard part of any cut-out is not the middle of the subject, it is the boundary, where a strand of hair is thinner than a pixel and genuinely half background. Edge softness spreads the mask over a few pixels so the outline does not look cut with scissors; two pixels suits most photos, zero gives a hard edge for keying, and six or more helps on fur and flyaway hair. Precise edges, which is on by default, does two further things in that uncertain band. It reads the actual colours around each pixel to work out how much subject and how much background are mixed into it, and sets the transparency from that measurement instead of from a blur. Then it unmixes the colour itself: a half-transparent hair pixel photographed against a white studio wall carries that white with it, and left alone it lights up as a pale rim the moment you drop the cut-out onto something dark. Undoing the mix is what removes the rim. Both steps need the subject and the background to differ in colour, since dark hair on a dark wall has nothing to measure, and both cost a few seconds on a large photo, which is what the Fast setting is there to skip.

Resolution, batches, and what the size limit is for

The longest side setting caps the output at 2048 pixels by default. That is not a licence limit, it is memory: the whole picture, its mask and several working copies have to fit in one browser tab at once, and a 24-megapixel photo multiplies quickly. Raise it to 4096 when you need print resolution and your machine has the room; lower it to 1024 if the tab runs out of memory. Twenty files can go in at once and come back as a ZIP. Each photo makes its own trip to the server as a 1024 × 1024 copy, so a batch of twenty uses twenty of the hourly server jobs. Photos that turn out to have no clear subject are still processed; you get the file and a note saying the cut-out came back nearly empty.

The original stays with you; our server sees only a 1024 × 1024 copy

To find the subject, the page shrinks the frame to a 1024 × 1024 copy and sends only that copy to our server over an encrypted connection; the server returns the mask. The copy is deleted as soon as the mask is ready, and the mask within about two hours. The original file is not uploaded: the cut-out, the edges, the background colour and the PNG encoding are made in this tab from the photo itself. It works the same way on a computer, a phone and a tablet. There is no account; each photo counts towards the shared hourly limit of the server tools. The result is written from raw pixels, which also drops the EXIF block: camera model, timestamp and GPS coordinates do not travel into the file you download.

Frequently asked questions

Is my photo uploaded to a server?

Only a reduced copy, not the photo. To find the subject, the page shrinks the frame to 1024 × 1024 pixels, sends that copy to our server over an encrypted connection and gets back only the mask, the outline of the subject. Your original file stays on your device: decoding, the cut-out, the edges, the background colour and the PNG encoding all happen inside this browser tab. The copy is deleted as soon as the mask is ready, and the mask within about two hours. It works this way on every device, computers included.

How long does one photo take?

Usually a few seconds for the mask, plus a few more for Precise edges on a large photo. Nothing has to be downloaded into the browser first, so the first photo is as quick as the ones after it. The server works out masks one at a time, so when many people use the tool at once a photo can wait a little in the queue.

Why does a server find the subject, not my browser?

Because the network is heavy. Run inside a browser tab, it first has to be downloaded and then holds more than 2 GB of memory while it works: a phone closes the tab, and many computers take half a minute per photo or fail on some graphics cards. On our server the same network returns the mask in a few seconds on any device, and your browser downloads nothing. The cost is a reduced 1024 × 1024 copy of the frame travelling over the network, which is why the note under the settings says so.

Is there a limit on how many photos I can do?

Yes, one. Each photo is a request to our server and counts towards the shared hourly limit of the server tools: 30 jobs an hour. Changing the background colour or the edges of the photo you just processed reuses its mask and does not count again. There is no sign-up, no watermark and no daily quota. Twenty photos per batch is a separate limit, set by the memory of the browser tab.

How good are the edges around hair?

Good, not perfect. Individual strands against a plain, contrasting background come out convincingly; hair against a busy background of similar colour is where every tool in this category struggles, including the paid ones. Precise edges is on by default and does most of the work; it is what stops a portrait shot on white from carrying a pale rim onto a dark backdrop. Widening Edge softness hides what is left. Switch to Fast if you would rather have the seconds back than the last of the fringe.

Can I process a whole batch at once?

Yes, up to 20 files, all with the same settings, returned as a ZIP. Each photo goes to the server as its own 1024 × 1024 copy, so a batch of twenty uses twenty of the hourly server jobs. If one file cannot be read, the rest still come back and you get a note saying how many were left out.

How is this different from Transparent Background?

Transparent Background removes one colour that you point at. It is the right tool for a logo or a scan on a flat white sheet, it needs no model and it runs instantly. This tool understands what a subject is, so it works on real photographs where the background is a room, a street or a sky. Use the simple one for flat backgrounds and this one for photographs.

Which formats can I put in?

JPG, PNG and WebP. HEIC and AVIF photos need converting first, and the tool will point you at the converter with your file already loaded. Format is detected from the file's contents rather than its name, so a mislabelled file is recognised for what it really is.

Guides

Related tools

Transparent BackgroundPick the background colour and everything close to it turns transparent: green screens, white backdrops, scans and flat logos, in a blink.Circle CropCut a circle, an oval or a rounded square out of any photo and download it as a PNG with a genuinely transparent background.Compress ImageShrink JPG, PNG, WebP, AVIF and GIF files in your browser: pick a level that keeps the resolution, or, for all but GIF, a target size in KB or MB.