Free Image Background Remover

Drop in a photo, a logo or a product shot and this free image background remover cuts the subject out for you, then hands it back as a transparent PNG. The model runs inside your browser, so the picture never gets uploaded anywhere. Files up to 25 MB.

Part of Piknova's free toolkit
photos, logos and products up to 25 MB

Your Photo Never
Leaves Your Laptop.

Every other free background remover works the same way: you hand them your image, their server does the thinking, and you take it on faith that the file gets deleted afterwards. This one does the thinking on your machine. The model downloads to your browser once, about 40 MB, and after that you could unplug the wifi and keep cutting out backgrounds on a train.

Free Forever Nothing Uploaded No Watermark

How to remove a background from an image

Three steps, no account, 25 MB per file.

1

Select your image

Click the button or drop a JPG, PNG or WebP onto the box. It stays on your device.

2

Press Remove background

A progress bar tells you what it is doing. The first run downloads the model; after that it is seconds.

3

Adjust and download

The cut-out opens in an editor. Pick a background, tidy the edges if you want, then choose your format and size.

The usual reason people are here

Make an image background transparent

Leave the background set to Transparent and the download is a PNG with a real alpha channel, not a picture of a checkerboard. Drop it on a coloured header, a slide, a mug mockup, anything, and it sits there with nothing behind it.

The checkerboard you see in the preview is the tool telling you the truth. A cut-out on a white card and a cut-out on a white background look identical on screen, so without something behind it there is no way to know which one you are about to download. The squares are how you check before you commit rather than after you have already dropped it into a design.

  • Partial transparency is kept: The edge of a subject is not a hard on-or-off line. Antialiased pixels, wisps of hair, the soft rim of a glass: those come out semi-transparent, which is what stops a cut-out from looking like it was scissored out of a magazine.
  • PNG, because JPG cannot: JPEG has no alpha channel at all. A transparent JPG is not a thing, so a transparent result always comes back as a PNG, and the tool says so before you download rather than after.
  • WebP keeps its own format: WebP does have transparency, so a WebP goes in and a WebP comes out. No format change, no size penalty for the privilege.

If you need the transparent version smaller, run it through the image compressor afterwards. A PNG with a lot of soft edges can be heavier than the photo you started with, which surprises people the first time.

Remove the background from a logo

Logos are the easiest thing this tool does and the thing people most often need it for. A flat mark on a flat background is exactly the shape a segmentation model is good at, so a logo usually comes out clean on the first pass.

The specific job is nearly always the same one: somebody sent you the logo as a JPG on white, and you need to put it on a dark header, a coloured banner, or a client's deck. Set the background to Transparent, remove, download the PNG. It will sit on any colour without the white box around it.

  • Text stays sharp: Wordmarks and taglines keep their edges, because the model works on the shape it finds rather than on a colour threshold. A colour-based eraser is what turns the counters inside letters like 'o' and 'e' into white blobs.
  • White inside the logo survives: This is the one that catches out magic-wand tools. Removing 'the white' removes every white pixel, including the ones that are part of the mark. Removing 'the background' does not.
  • Full resolution comes back: A logo that goes in at 2000px comes out at 2000px. Nothing is downscaled to sell you an upgrade.

Honest limit: this returns a PNG, which is pixels. If your logo needs to scale to the side of a van, you want the original vector from whoever designed it, and no background remover on the internet can hand you that.

For sellers

Product photos for Amazon, Etsy and eBay

Marketplaces are picky about backgrounds in a very specific way, and the requirement is usually the same three words: pure white, RGB 255,255,255. That is what the White option here is, exactly, rather than an off-white that looks fine on your monitor and gets the listing flagged.

Where it is goingWhat they ask forWhat to pick here
Amazon main imagePure white background, product filling most of the frameWhite, then crop and resize afterwards if you need to
Etsy listingNo hard rule, but consistent backgrounds sell betterWhite for a clean grid, or a brand colour for a themed shop
eBayWhite or very light background on the gallery imageWhite
Your own store or a mockupWhatever your template needs behind itTransparent, so the template decides
Instagram or an adSomething that stands out in a feedPick a colour, and use the same one across the set
Marketplace rules change and this table is a summary, not a legal document. Check the current policy for the site you are listing on before a bulk upload.

The practical workflow for a batch of listings is boring and it works: shoot everything against whatever you have, cut them all out on White here, then resize the set to the same dimensions with the image resizer. Consistency across a grid of thumbnails does more for a shop than any single photo does.

Removing a white or a black background

These two get searched for constantly and they mean something slightly different from what most tools do about them. Worth thirty seconds, because the difference is why colour-based erasers keep eating parts of your subject.

A classic background eraser works on colour. You click the white, it deletes everything close to white, and it does not know or care whether those pixels were behind your subject or inside it. That is why the highlight on a white mug vanishes, why the whites of eyes go see-through, and why a logo with white letters comes back full of holes.

This one works on the subject. It looks at the picture, decides which pixels belong to the thing in front, and keeps those regardless of what colour they happen to be. So a white product on a white background works, a black cat on a black sofa works, and neither one needs you to fiddle with a tolerance slider.

  • White backgrounds: Studio shots, scanned documents, and anything that came off a marketplace listing. Nearly always a clean result.
  • Black backgrounds: Product shots on dark surfaces, stage photos, thumbnails. Same treatment, same one click.
  • Busy backgrounds: A street, a room, a crowd. Harder, and where the model earns its download, because there is no single colour to key out at all.

If what you actually want is to swap a white background for a black one, do it in one step: remove, then pick Black. There is no reason to download a transparent PNG and go and paste it onto something else.

JPG, PNG and WebP: what you get back

The rule is simple and the tool tells you before you download: you keep the format you brought, unless that format physically cannot hold the result.

You uploadYou pickYou get
JPG or JPEGTransparentPNG, because JPEG has no alpha channel to store transparency in
JPG or JPEGA solid colourJPG, unchanged in type
PNGAnythingPNG
WebPAnythingWebP, transparency included
The only format change that ever happens is JPG to PNG for a transparent result, and it is flagged next to the background chips before you press anything.

This matters more than it sounds. The lazy way to handle a transparent result out of a JPG is to paint white behind it and hand back a JPG anyway, which looks correct in a thumbnail and then turns up as a white box the moment the image lands on a coloured background. We would rather change the extension and say so.

Headshots, profile pictures, passport photos and signatures

People are what this model was trained hardest on, so portraits are where it looks most impressive and where its remaining flaws are easiest to spot.

  • Headshots and profile pictures: Cut the person out, drop them on white or a brand colour, and a photo taken in a kitchen becomes usable on a team page. Shoulders and collars come out cleanly; flyaway hair is the part worth zooming in on before you commit.
  • Passport and ID photos: Most of these want a plain white or light background. Cut out, pick White, then use the image resizer for the exact dimensions the form demands and the compressor for its file size limit.
  • Signatures: Photograph or scan the signature on paper, remove the background, and you get the ink on transparent with the paper texture and the shadow gone. That is a genuinely useful trick for filling in PDFs.

Two honest cautions. Some official ID photo rules require an unedited photograph, so check before you submit a cut-out to a government form. And for signatures, remember that a transparent PNG of your signature is worth treating like a password, which is a decent argument for using a tool that never uploaded it in the first place.

An automatic background remover, not a background eraser

Both phrases get typed into Google roughly the same amount, and they describe two quite different tools. This is the first kind.

  • A background eraser: You do the work. Brush, lasso, magic wand, tolerance slider, then twenty minutes of fixing the bits it took too much of. Full control, and full responsibility.
  • An automatic background remover: The model does the work. One click, a few seconds, and a result that is either very good or occasionally wrong in a way you cannot nudge. No control, no twenty minutes.

For the overwhelming majority of jobs, the automatic one wins on the only axis that matters, which is that the job actually gets done. Nobody hand-masks forty product photos. The exception is a single hero image where the edge has to be perfect, and for that you want a real editor with a brush, not a browser tab.

You get both here, in that order. The model does the work, and the result arrives in an editor rather than as a thumbnail with a download button: shrink the edge to kill a leftover halo, erase a patch of background it kept, brush back a piece of the subject it cut off. The automatic pass does ninety percent of the job in a few seconds, and the brush is there for the last ten percent instead of for the whole thing.

The part most free tools skip

The result opens as an editor, not a thumbnail

Every background remover looks perfect at thumbnail size, which is exactly why this one does not hand you a thumbnail. The moment the cut-out is ready it fills the workspace with the tools around it, so checking the edges is the next thing you do rather than something you have to go looking for.

  • Shrink edges: Pulls the cut-out inwards a pixel or two. This is the fix for a halo, which is the single most common flaw in an automatic cut-out: the pixels right on the boundary genuinely were a blend of subject and background, and the model had to round them one way or the other.
  • Soften edges: Blurs the boundary slightly so a shrunk edge stops looking like it was cut out with scissors. A pixel or two is usually the difference between a composite that reads as real and one that reads as a sticker.
  • Erase: Paint away a patch of background the model decided to keep. Shadows under a product and reflections in glass are the usual culprits.
  • Restore: Paint back a piece of the subject it cut off. A pale sleeve against a pale wall, the handle of a mug, the leg of a chair.

Everything in that view edits the transparency, never the picture, which has a consequence worth knowing: after you have refined an edge you can still switch from transparent to white to a brand colour and the work follows. Undo, redo and reset are all there, and reset really does go back to the model's original cut-out rather than to your last save.

The preview is scaled down for images past about 2800 pixels on the long edge, so a phone does not run out of memory holding a full-size editable copy. It says so when it happens, and the file you download is always built at full resolution from the edits you made.

The trade we made

Why this one downloads 40 MB instead of uploading your photo

Every background remover has to run a neural network somewhere. The choice is whose computer it runs on, and that choice is the whole difference between this tool and the rest of the results on this page.

The usual arrangement is that you upload the image, a server segments it, and you download the result. It is fast, it needs nothing from your machine, and it means your product shots, your ID photo and your client's unreleased logo all sat on somebody else's disk for a while. Most of those services are perfectly honest about deleting them. You still have to take their word for it.

Here the model comes to you instead. The first time you press the button, your browser fetches the segmentation model and its runtime, about 40 MB, and caches it. Every run after that is local, and the image itself is never transmitted at all. Not to us, not to anyone. There is nothing to delete because there was never a copy.

  • What this costs you: A one-time 40 MB download, and a few seconds per image instead of one, because your laptop is not a datacentre. On mobile data that first run is a real cost and we would rather say so than let you find out from your bill.
  • What it buys you: No upload, no account, no credits, no daily cap, no watermark, and no resolution tier. Your device does the work, so a hundred cut-outs cost us exactly what zero do.
  • What still touches the network: The model files themselves, which come from a CDN like any other script on the internet. That is a download to your browser, not an upload of your picture, and it is the only network traffic this tool generates.

The honest ceiling: browsers run out of memory eventually. 25 MB and 25 megapixels per image is where we stop, because past that a full-resolution cut-out can take a phone's tab down with no useful error. Desktop software still wins on giant batches and on RAW.

Everything the paid ones charge for, and one thing they cannot sell.

Full resolution, no watermark, no credits, and a file that never reaches a server.

One click, no brush required

A segmentation model finds the subject and traces it. You do not paint a mask, drag a lasso, or clean up the corners. Drop the file, press the button, look at the result.

Transparent, white, black or your brand colour

Pick what goes behind the subject before you download. Marketplace listings want pure white, thumbnails usually want something darker, and a logo wants nothing at all.

Change the background without running it again

The cut-out is kept once it is made, so swapping from transparent to white to a blue you picked is a repaint rather than a second wait. Try all four in about a second.

The result opens as an editor, not a thumbnail

When the cut-out lands it fills the workspace at full size, with the tools around it. Zoom in on an edge, shrink or soften it, erase a patch the model kept, brush back a piece it cut off, and hit Before at any time to see what you started with.

Nothing leaves your device

The model comes to your image instead of the other way round. Your file is read from disk, processed in memory, and handed back. There is no server to trust, which is a stronger promise than any privacy policy, ours included.

Full resolution, every time

You get the same pixel dimensions you uploaded, at no charge, with nothing stamped across the corner. There is no HD tier here because there is nothing for us to hold back.

Works on your phone

Same tool, same result on mobile. Cut out a product straight from your camera roll and list it without opening a laptop, though the first run is a chunky download so wifi is kinder than mobile data.

Frequently asked questions

How the background remover works, and where it stops.

Drop your JPG, PNG or WebP onto the box at the top of this page, choose what you want behind the subject, and press Remove background. A few seconds later you get a preview with the original next to the result, and a download button under it.

There is no account, no credit system and no watermark. The first run downloads the model to your browser, which takes a moment; every run after that starts immediately.

No. The segmentation model runs inside your browser, on your own hardware. Your file is read from your device, processed in memory, and handed back. Nothing is transmitted, so nothing can be stored, logged or leaked, and closing the tab erases every trace.

The one thing that does travel over the network is the model itself, which your browser fetches from a CDN the first time you use the tool and then caches. That is a download to you, not an upload of your picture.

Because your browser is fetching roughly 40 MB of neural network before it can do anything with it. That is the price of not sending your image to a server, and it is paid once: the files are cached afterwards, so the second image starts instantly.

If you are on mobile data and the download matters, wait for wifi. The progress bar tells you which phase you are in so you are never staring at an unlabelled spinner wondering whether it has hung.

Yes, and that is the default. Leave the background on Transparent and you get a PNG with a real alpha channel, including the partial transparency along antialiased edges, so it sits cleanly on any colour you drop it onto.

If you uploaded a JPG, the download comes back as a PNG, because JPEG cannot store transparency at all. The tool flags that next to the background options before you press anything.

Logos and products are the easiest cases, and usually come out clean on the first pass. Text keeps its edges, and white areas inside a mark survive, which is the thing a colour-based eraser gets wrong.

For a marketplace listing, pick White rather than Transparent. Amazon and eBay want pure white behind the product, and the White option here is exactly RGB 255,255,255 rather than an off-white that looks right and gets flagged.

If you choose a solid background, yes. A JPG comes back as a JPG, a PNG as a PNG, a WebP as a WebP. Changing your file's type without being asked is a good way to break somebody's upload form.

The single exception is a transparent result from a JPG, which has to become a PNG because JPEG has no alpha channel. The alternative would be painting a white background behind your subject and calling it transparent, which is how a logo ends up in a white box on a coloured header.

Better than you expect and not as good as a person with an hour. Individual strands of hair against a clean background usually survive with their soft edges intact. Hair against a busy background of a similar colour is where it starts guessing, and sometimes guesses wrong.

Zoom in and see for yourself, because the result opens in an editor rather than as a preview. There is a shrink slider for a leftover halo and erase and restore brushes for anything the model got plainly wrong, so a nearly-right cut-out is a thirty second fix rather than a reason to start again somewhere else.

What a brush cannot do is invent detail the model never captured. If a strand of hair was blended into a background of the same colour, painting it back gives you a blob in the shape of your brush. For a hero image at that level, the honest answer is still a different source photo or a person with an hour.

Yes, and the tools are already on screen when the cut-out arrives. There is a shrink slider for a leftover halo, a soften slider for a hard edge, and erase and restore brushes for anything the model plainly got wrong. Change the background while you work, because a fringe that is invisible against one colour is obvious against another.

The edits change the transparency rather than the picture, so you can still swap the background afterwards without losing them. Undo, redo and a reset back to the original cut-out are all there.

PNG, JPG or WebP, at original size or scaled down to roughly 2000, 1200 or 600 pixels on the longest edge. The Download button opens both lists, and it only offers sizes that are actually smaller than your image, because a menu that offers to enlarge something it cannot enlarge is just a broken row.

Every combination is free. There is no HD tier and no watermark on the small ones. The one option that can be unavailable is JPG while the background is transparent, because JPEG has no alpha channel; pick a background colour and it comes back.

No. No daily cap, no credits, no account, no watermark, and no resolution tier. Your device does the work, so there is nothing for us to ration.

One image at a time for now. Batch removal is on the list, and it is the feature that would actually change this tool's life for anyone running a shop.

25 MB per image, with a 25 megapixel ceiling on top of it. That covers phone and DSLR photos comfortably.

The pixel limit is lower here than on the other tools because this one holds the decoded image, a full-size mask and an RGBA composite in memory at once. Past that a browser can run out of memory in a way that shows up as a blank tab rather than an error, so the tool declines the file instead of gambling with your work.

Remove a background now

Drop in an image, pick what goes behind it, download the result. Nothing gets uploaded.