Remove an image background

Your framing is kept, not "improved"

Tools built for product listings crop to the outline of the subject and recentre it, because a marketplace demands that. If you only wanted a transparent PNG, that is a surprise you then have to undo. Here the image comes back with the same dimensions and the same composition you dropped. Only the background is gone. Subject off to the left stays off to the left.

Want a new size as well? Fill in a width or a height and the whole frame is scaled, still without recentring anything.

No queue, no credits, no upload

Every hosted background remover meters you: a few free cut-outs, then a plan, and a full-resolution download behind a paywall. That model exists because the segmentation runs on their servers and their servers cost money. Move the model to your machine and the whole meter disappears, which is what happens here.

The first time you switch the option on, the weights are fetched once from the Hugging Face CDN and cached by your browser. After that it starts instantly and keeps working with no network at all. Your image is never part of any request.

Which model you get, and why it matters

The cut-out is done by BiRefNet (MIT licence). Two builds ship, and the right one is picked for your hardware without asking you:

Your GPU's shader limits are checked before choosing, so the tool drops to the CPU build up front rather than failing halfway through a batch. The download bar reports real bytes received, not a fake animation.

How to remove a background from an image

1. Drop your image. JPEG, PNG or WebP, up to 25 MB, up to 50 at a time. The model loads on the first drop and is reused afterwards.
2. Choose what replaces the background. Transparent gives you a PNG you can put on anything. White gives you a much lighter JPEG. Add a width or a height if you also need a different size.
3. Download the cut-out. Full resolution, no watermark, one by one or as a single zip.

Frequently asked questions

How does an AI background remover work?

A segmentation network looks at the image and predicts, for every pixel, how likely it is to belong to the subject rather than the scene. That prediction becomes the alpha channel of the output, which is why a good one handles hair and thin edges rather than cutting a hard outline. Here that network is BiRefNet, and it runs in your browser instead of on a server.

Can I remove a background without uploading the photo anywhere?

That is the entire point of this page. The model is downloaded to you; your photo stays put. Nothing about it is sent, logged or stored, and no account exists to attach it to.

Will it keep the original resolution?

Yes. The cut-out is applied at full resolution and the output keeps the source dimensions unless you explicitly ask for a different size. There is no downscaled free tier.

Is it really free, with no watermark?

Yes. No account, no trial, no export limit, nothing stamped on the image. Discreet ads pay for the hosting.

Can I do a whole folder at once?

Up to 50 images per batch, processed one after another, downloadable as one zip. The bulk page explains what a large batch costs in time and memory.

Why does the first cut-out take longer than the rest?

Because the model is being downloaded, once. A progress bar shows it. Every batch after that starts immediately, including offline.