Skip to content
Remove Bg Picture

About Remove Bg Picture

Remove Bg Picture is a free, browser-based tool that removes the background from pictures using a neural network that runs on your own device. This page covers what it does, how the AI works technically, and exactly what happens (and does not happen) to your files.

What it does

Drop one picture or up to 15 at once, and the tool detects the subject in each photo and makes everything else transparent. The result is a PNG at the original resolution with a true alpha channel — ready for product listings, profile pictures, presentations, thumbnails, and collages, or for placing on a new background in any editor.

A before/after slider lets you inspect the cutout against the original, and batch results can be downloaded individually or all at once. There is no watermark, no signup, and no usage cap.

How it works technically

The tool runs U²-Net (Qin et al., Apache 2.0), a salient-object-detection neural network widely used for automatic background removal, in its lightweight “u2netp” variant — a 4.5 MB ONNX model. Inference happens directly in your browser through ONNX Runtime Web compiled to WebAssembly with SIMD. The pre- and post-processing follows the same steps as the popular open-source rembg project: the picture is scaled to 320×320 and normalised, the network predicts a saliency map, and that map is min-max normalised, upscaled back to the original resolution, and applied as the alpha channel.

The model file and the WebAssembly runtime are served from this site itself and cached by your browser, so after the first visit the tool loads instantly and works offline. No third-party AI API is involved — there is no per-image cost, which is why the tool can stay free.

Automatic segmentation has limits: extremely fine hair, glass, and semi-transparent objects are hard for any model, and the lightweight variant trades a little edge precision for a much faster download. For most photos — products, people, pets, objects — the cutout is clean and immediately usable.

Privacy: your pictures stay on your device

Everything — decoding, AI inference, compositing, PNG encoding — happens locally in your browser. Your pictures are never uploaded, stored, or analysed on a server. We could not look at them even if we wanted to, because they never reach us.

Once the page and the model have loaded, you can disconnect from the internet entirely and the tool keeps working. The site uses ordinary analytics and advertising scripts to stay free, but those never see your files.

Credits and licences

U²-Net was created by Xuebin Qin and co-authors and is distributed under the Apache 2.0 licence. The ONNX export of the u2netp checkpoint originates from the rembg project (MIT). ONNX Runtime is an open-source project by Microsoft (MIT). This site bundles these components unmodified and credits them with thanks.

Have feedback or found a bug? Reach out on Twitter/X, or try our related tool Compress Picture for shrinking image file sizes.