Image

Remove the background from an image

Identifies the main subject of a photo and erases the background around it, running a segmentation model entirely in your browser.

Drag one or more images here, or click to choose

JPG, PNG, or WebP, up to 25 MB each

Output

Transparent backgroundPNG — keeps the cutoutSolid color backgroundpick the color belowMask onlyblack and white, no color

Format: PNG — it's the only one that stores the cutout's transparency, so there's no format option here.

The model is good at well-defined silhouettes (people, products, solid objects) and struggles more with loose hair, animal fur, sheer fabric, glass, and smoke. Check the result before using it — especially around the edges.

Result

The result appears here after processing.

Segmentation model: ormbg (Apache License 2.0). Runs entirely in your browser — the file is never uploaded to any server.

Need to manually pick the area to keep, instead of removing the background automatically? Use crop image. To just change the file format, without touching the content, use convert image.

Local processing: your file never leaves your computer.

Does:Identifies the main subject of a photo and erases the background around it, running a segmentation model entirely in your browser.
Processes:Your image is processed entirely in your browser — nothing is uploaded to a server.
Doesn't do:The model used (ormbg, Apache 2.0 license) and the engine that runs inference (onnxruntime-web, MIT license) use WebGPU when the browser supports it, with a slower mode (WASM) as an automatic fallback — the files are hosted on the same origin as the site, with no call to a third-party server.
First use:Downloads a large file (a model or engine) from the site’s own server the first time you use it — the file you process still never leaves your browser. See "How it works" on the home page.

Cutting out a background without sending the file anywhere

Separating a photo's subject from whatever is behind it — a product for a catalog, a headshot for a badge, an object you want to paste onto a different background — usually means opening an image editor and spending some time with a manual selection tool. This tool does that separation automatically: a segmentation model analyzes the image, figures out what looks like the main subject, and erases everything around it.

What sets this tool apart from a typical background-removal service is where the processing happens: entirely inside your browser, with a model that runs locally. The image you upload is never transmitted to any server — there's no upload, because there's nowhere for it to go. The trade-off is an initial download: the model that does this recognition is a large file, quite different from the rest of this site's tools, which use only lightweight code.

A download that only happens once

Before processing your first image, the tool tells you it's about to download roughly 68 MB — the segmentation model plus the engine that runs it in your browser — and asks for explicit confirmation before starting. That download only happens this first time: your browser caches the files, and future uses, even after closing the tab or restarting your computer, start processing right away with no re-download. That's why the warning shows up once per browser, not once per image.

Processing itself uses WebGPU when your browser supports it, which makes analyzing each image noticeably faster. Without WebGPU, the tool automatically falls back to a slower mode (WebAssembly) — no configuration needed, but the interface tells you when that's the case, so a longer wait makes sense.

Three ways to get the result

By default, the result comes back as a PNG with a transparent background — the format that preserves the cutout so you can place it over anything else afterward. There's also an option to deliver the result against a solid color you choose, which opens up exporting as JPG or WebP (formats with no transparency, so they only make sense with a background already filled in). A third option delivers just the mask: a black-and-white image marking what was identified as subject and background, useful if you're going to use the cutout as a reference inside another program.

Where the cutout tends to fail

No segmentation model gets it right every time, and it's worth knowing the common failure cases before trusting a result without checking it. Well-defined outlines — a person against a plain wall, a product on a studio backdrop — usually come out clean. Loose hair, animal fur, sheer fabric, glass, and smoke are notoriously harder: the line between "subject" and "background" in those textures is ambiguous even to the human eye, so the model struggles there too. That's why the preview can be checked against a checkered background, a dark background, and a light background — a subtle edge flaw sometimes only becomes visible depending on the contrast behind it.

Processing several images at once

When you upload more than one image, each goes through the same process individually, in sequence — not in parallel, since a single analysis already uses a good amount of processing power on its own. This is the heaviest processing job on the site, so large batches can take minutes to finish, and the tab needs to stay open the whole time. Cancelling stops the batch at any point, and each image's memory is released as soon as it's done processing, before the next one starts. The result comes back as an individual download per image, or a single ZIP with the whole batch.

What this tool doesn't do

It doesn't edit the image's content beyond cutting out the background — it doesn't retouch, doesn't remove unwanted objects from within the subject, doesn't sharpen or adjust color. It doesn't replace a manual cutout when outline precision matters more than automation can deliver — for choosing the area by hand, a cropping tool is a better fit. And despite the model's care, it doesn't get it right every time — the result needs to be checked, especially at the edges, before you consider the cutout ready to use.

Frequently asked questions

Do I need to install anything or create an account to use this?

No. The tool downloads a segmentation model (about 68 MB, the first time you use it) and runs everything inside your browser — the image you upload never leaves your computer, and there's no upload to any server.

Why do I need to confirm a download before using the tool?

Because the model that recognizes the subject of a photo is a large file, unlike the rest of this site's tools, which process everything with lightweight code. That download only happens the first time: once it's cached, your browser stores the files locally, and future uses start processing right away, with no re-download.

When does the cutout tend to come out wrong?

Well-defined outlines — a person, a product, a solid object against a simple background — usually come out clean. Loose hair, animal fur, sheer fabric, glass, and smoke are the hardest cases: the edge in those areas is inherently ambiguous, and the model makes more mistakes there. It's always worth checking the cutout's edge before using the result, especially in those cases.

What's the difference between a transparent background, a solid background, and mask only?

Transparent background gives you a PNG with the subject cut out and everything around it transparent, ready to place over anything else. Solid background replaces what was removed with a flat color you choose, and accepts JPG, PNG, or WebP. Mask only gives you a black-and-white image showing what was identified as subject (white) and background (black) — useful if you're going to use the cutout as a reference in another program.

Does processing several images at once take a long time?

Yes, proportionally — this is the heaviest processing job on the site, because each image goes through a full segmentation model, not just a pixel adjustment. A large batch can take minutes, and the tab needs to stay open until it finishes. Images are processed one at a time, and you can cancel at any point.