Background Remover

Remove the background from any photo with AI that runs in your browser: no upload, transparent PNG.

AI model

Checking what your browser supports…

Background

A PNG with a transparent background, ready for any design.

Applies to the previews and the downloads. Images are saved as PNG at their original size.

Drop photos here or click to choose files JPG, PNG or WebP, one or several: they are processed one after another. You can also paste an image with Ctrl+V.

Embed this tool on your site

Free and without an account: copy the code and paste it into your website, blog or docs.

About the Background Remover

Remove the background from a photo without sending it anywhere. The cut-out is made by an AI model that runs on your own device: the first time, your browser downloads it from Hugging Face and keeps it, so later visits start right away. Drop one photo or several, compare each result with the original, and download a transparent PNG at full size, or put the subject on a solid color or a blurred background.

How to use the Background Remover

  1. Load the model

    The AI model card shows which model your browser will use, its license and its download size: about 123 MB with WebGPU, about 44 MB without it. Click Download the model, or simply add an image and the download starts by itself. If it was downloaded on an earlier visit, the card says Ready, already downloaded.

  2. Add your photos

    Drop JPG, PNG or WebP images on the page, click to choose them or paste one with Ctrl+V. They are processed one after another, and each one shows the original and the result with a slider you can drag to compare them.

  3. Choose the background and download

    Pick Transparent, Color (with ready-made colors or your own) or Blur in the Background panel; it applies to the previews and to the downloads. Click PNG to save an image at its original size, or Download all (.zip) to get every result at once.

When to use it

Product photos for online stores

Put products on a clean white or colored background for a marketplace listing or a catalog, or keep them transparent to place them in a banner, without paying per image or uploading unreleased products.

Profile pictures and portraits

Cut yourself out of a casual photo for a profile picture, a CV or a team page, on a solid color that matches your brand or with the original background blurred like portrait mode.

Designs, slides and stickers

Get a transparent PNG of a person, a pet or an object to use in a presentation, a thumbnail, a poster or a sticker, ready to drop into any design tool.

Frequently asked questions

Are my images uploaded?

No. The AI model runs in your browser, and the images are decoded, cut out and saved as PNG on your device; nothing is sent to a server, so it also suits confidential photos. The internet is used only to download the model from Hugging Face and the AI engine (Transformers.js and ONNX Runtime, about 6 MB) from jsDelivr the first time. Only your background settings are remembered in this browser.

Which AI model does it use, and under what license?

With WebGPU it uses BiRefNet lite, by Peng Zheng and colleagues, MIT license, in an ONNX version adapted to run on WebGPU (about 123 MB). Without WebGPU it uses ORMBG, based on IS-Net, Apache-2.0 license (about 44 MB, 8-bit), because BiRefNet needs more memory than WebAssembly allows. Both licenses allow commercial use, and the page names the model and links its card on Hugging Face before anything is downloaded.

Do I have to download the model every time?

No. Your browser keeps the model after the first download, and the card then says Ready, already downloaded. It is downloaded again only if you clear the site data or your browser frees up space. Once the model and the engine are stored, the tool also works offline.

Why is it faster on some devices?

With WebGPU (recent Chrome and Edge, and other browsers that support it) the model runs on your graphics card and an image takes a few seconds; the first one takes longer while the model is prepared. Without WebGPU it runs on the processor with WebAssembly and takes about 10 to 30 seconds per image. If WebGPU fails, the tool switches to WebAssembly by itself and tells you.

What works best, and what does not?

It works best with a clear subject that stands out from the background: people, pets, products, cars. Hair and fur keep soft edges, but fine strands against a busy background can lose detail or keep a faint halo, and the lighter model used without WebGPU can leave traces of the background. Glass and other transparent objects keep the old background showing through them, and shadows on the floor are removed. The result is always a PNG at the original resolution.