🖼 ConverterWebP
Free • No sign-up • No watermark

Remove image background

Erase the background and download a transparent PNG. One mode removes the background, the other isolates the subject — both running in your browser.

Drag and drop your images here

or click to select from your device

JPG, PNG, WebP, AVIF • up to 20 images • output is always PNG

🔒

100% Secure: Your photos are processed in your own browser and never pass through any server.

Fast mode

Removes the background and keeps everything else in the picture. Instant, nothing to download. Handles a flat colour or a smooth gradient.

32
StrictAggressive

Raise it if patches of background survive; lower it if parts of your subject disappear.

🧠

AI mode

Keeps the main subject and discards everything else — including things you may have wanted. Reads hair, fur and busy scenes the fast mode cannot.

First use downloads about 50 MB (the segmentation model plus its runtime), fetched from a third-party CDN and then cached by your browser. Your image is still never uploaded — only the model comes down. On mobile data, prefer fast mode.

⚡ Instant on solid backgrounds

A logo on white is done before you finish reading this sentence.

🎨 Transparent PNG out

Drop the result onto any background in Canva, Figma, Word or Photoshop.

🔒 Your photo stays put

Neither mode uploads the image. AI mode downloads a model — never your file.

How background removal works, and which mode you actually need

"Remove the background" describes two genuinely different problems, and most tools quietly solve only one of them. Cutting a logo out of a white rectangle is a matter of arithmetic. Separating a person's windblown hair from a park behind them is a matter of judgement. This page offers a mode for each, and picking the right one is most of the battle.

Fast mode: what "uniform background" really means

Fast mode starts at the outer edge of your image and floods inward, marking every pixel that is close enough in colour to where it started, and stopping when the colour changes sharply. That is why it is instantaneous and needs nothing downloaded — it is comparing numbers, not recognising objects.

It works beautifully on a logo on white, a screenshot, a scanned signature, a product shot taken against a seamless backdrop — and, because each step compares a pixel with its neighbour rather than with the starting point, on smooth gradients too: a sky, a wall with a shadow falling across it, a studio backdrop with a vignette. It has two failure modes worth knowing. First, it only removes background connected to the edge: the enclosed gap inside the letter "O" or between an arm and a torso stays filled, because the flood never reaches it. Second, anything in your subject that matches the background survives or vanishes with it — a white shirt on a white background will be eaten through. The tolerance slider is the dial between those two failures: too low leaves a halo of background, too high starts eating the subject.

The two modes answer different questions

This is the distinction that should decide which one you pick, and it is not "easy background" versus "hard background". Take a photo of a bird perched on a branch against the sky. Fast mode removes the sky and leaves the branch: it follows colour spreading in from the edge, and the branch is a different colour, so it stays. AI mode removes the sky and the branch: it never looks for background at all — it looks for the subject, decides the bird is it, and discards everything else.

So the question is not which one is more accurate. It is whether you are asking "take the background out" or "give me only the subject". A product photographed on a table: fast mode keeps the table, AI mode hands you the product alone. Neither is wrong, and picking the wrong one is the most common way to end up with a cut-out missing something you wanted. Technically, AI mode runs a segmentation model — a neural network that decides pixel by pixel what belongs to the subject — which is what lets it cope with strands of hair, fur, the gap between fingers or a translucent edge.

The cost is real and worth stating plainly. The model and its runtime are roughly 50 MB, downloaded from a third-party CDN the first time you use the mode, then kept in your browser cache. On a fast connection that is a few seconds; on mobile data it is a meaningful chunk of an allowance. Processing then takes a few seconds per image and is noticeably slower on older phones. Everything after the download happens locally — the model comes to your image, not the other way round.

Why the result is always a PNG

Transparency needs an alpha channel, and JPG does not have one. Save a cut-out as JPG and the transparent area collapses to a solid colour, which defeats the entire exercise. Both modes therefore output PNG, and the preview shows a chequerboard behind each result so you can see the transparency rather than guess at it. If you need a smaller file for the web afterwards, convert the PNG to WebP — it keeps the alpha channel.

Getting a clean cut

If you control how the photo is taken, you control most of the result. Put the subject well clear of the background so the two are not lit alike; avoid a backdrop that shares colours with the subject; keep the lighting even to stop shadows creeping across the backdrop; and shoot at the highest resolution you can, because a cut edge that looks rough at 500 pixels wide looks fine at 2000. For fast mode specifically, a plain sheet of paper or a wall is worth more than any slider.

Frequently asked questions

Ask what you want to keep. If the answer is "everything except the background", use fast mode: it is instant, downloads nothing, and copes with a flat colour or a smooth gradient such as a sky or a studio backdrop. If the answer is "only the subject", use AI mode — it isolates whatever it judges to be the main object and drops the rest, including the surface underneath it. Ragged patches from fast mode usually mean the tolerance needs adjusting, not that you need the AI.

Because it is downloading the model — around 50 MB of neural network weights and WebAssembly runtime — before it can process anything. That happens once: your browser caches the files, so the second image and every later visit start immediately. If the download fails, check that no network filter is blocking the CDN and fall back to fast mode meanwhile.

Not here — this tool produces the cut-out, and composing it over something else is a design job. The transparent PNG you download drops straight into Canva, Figma, Google Slides, PowerPoint, Word or any image editor, where you can place whatever background you like behind it.

No. Fast mode makes no network request at all. AI mode makes one kind of request — downloading the model files from a CDN — and that traffic goes one way: files come down to your device, your image never goes up. The neural network runs on your own processor through WebAssembly.