Remove Image Background

Choose a photo and the background is removed right here in your browser. Nothing is uploaded.

Drop a photo here or tap to choose You can also paste an image with Ctrl/Cmd+V

JPG, PNG, WebP, AVIF, GIF or BMP. HEIC photos from iPhones open only in browsers that can decode them (Safari). Images larger than 16 megapixels are scaled down to fit phone memory limits.

How it works

When you choose a photo, your browser downloads a small image-segmentation model (U²-Net-p, 3.4 MB, cached for next time) from this site and runs it on your own device in a background thread with WebAssembly. The model marks which pixels belong to the main subject; an edge-refinement step then aligns the mask with the photo's real edges, and the cut-out is assembled at full resolution in your browser.

Is my photo uploaded?

No. There is no server that receives images: the page is static and its security policy blocks connections to any other site. Your photo is decoded, analysed and exported entirely in your browser, and it is gone when you close the tab.

Tips for good results

The model works best when the subject is clear: a person, product, animal or object that stands out from the background. Busy scenes with several subjects, very thin details such as loose hair, and transparent objects like glass are harder. Use a colour or blurred background to check the edges before downloading.

Which formats are supported?

Anything your browser can open: JPG, PNG, WebP, AVIF, GIF and BMP. HEIC/HEIF photos from iPhones only open in Safari; on other browsers, export them as JPEG first (on iPhone: Settings, Camera, Formats, Most Compatible). The result downloads as a PNG with a transparent background, or with the colour or blur you picked.

Credits

Segmentation model: U²-Net-p by Xuebin Qin, Zichen Zhang, Chenyang Huang, Masood Dehghan, Osmar R. Zaiane and Martin Jagersand (“U²-Net: Going Deeper with Nested U-Structure for Salient Object Detection”, Pattern Recognition 2020), licensed under the Apache License 2.0 (licence and notice). Weights converted to our own format; the inference engine is Talisman's own code.