What usually happens to your photo
A typical online background remover works like this: your browser uploads the image to the company's servers, a model runs on their hardware, and the result is sent back. Many services then keep the original and the result for a period: hours, days or longer, depending on their privacy policy. Some use uploads to improve their models. For a product shot that may not matter. For a photo of your child, a scan of your signature, an ID document or anything under a confidentiality agreement, it might.
Running the model in the browser instead
Modern browsers can run neural networks themselves. WebAssembly runs compiled code at near-native speed, and WebGPU gives web pages access to the graphics card. ONNX Runtime Web, an open-source engine from Microsoft, uses both. This site's background remover downloads a compact model (MODNet for people at 6.6 MB, U²-Netp for objects at 4.6 MB, both Apache-2.0 licensed) and runs it locally. Your photo is decoded, segmented and re-encoded in the page's memory. When you close the tab it is gone.
There are trade-offs. The first run downloads the model, and an older phone takes a few seconds per image where a server would take one. The models are smaller than the largest server models, so very difficult edges, like fine hair against a busy background, may need a touch of the brush. For most photos the result is comparable, and the privacy difference is absolute.
How to prove nothing is uploaded
- Open the background remover in Chrome, Edge or Firefox on a computer.
- Press F12 to open the developer tools and switch to the Network tab. Tick Preserve log.
- Load a photo and remove its background. Watch the requests: you will see the page's own files, the model file from this site, and the ONNX Runtime WebAssembly file from jsDelivr. They are all downloads to your browser.
- Sort by size or filter by "Fetch/XHR": no request carries your image, and none is an upload of more than a few kilobytes.
- For a stronger test, after the first image, set Throttling to Offline and process another. It still works, because nothing needs the network.
On a phone, do the same with aeroplane mode once the model has loaded. The site's privacy proof page lists every host the site may contact.
When local processing matters most
- Signatures and documents. A clean signature image is exactly what fraudsters want. The signature mode here does not even use AI; it is plain image processing.
- Children's photos. School, sports and family photos are better kept off third-party servers.
- Client and unreleased work. Product shots under embargo, client portraits and design assets covered by an NDA.
- Workplace rules. Many employers forbid uploading work images to unapproved services. Local processing sidesteps the question.
Reading another tool's privacy policy
If you do use an online service, a few questions tell you most of what you need. Does the policy say images are processed on its servers, and for how long are they kept: deleted immediately, after 24 hours, or "as long as necessary"? Are uploads used to train or improve models, and can you opt out? Which country are the servers in, and which third parties (cloud hosts, analytics, payment processors) can see the files? Is there a way to delete your images on request? A service that runs in the browser can answer all of these with "your image never reaches us", which is easier to trust because you can check it yourself with the steps above.
What the free version does
Everything: all modes, the brush, every background option, with no watermark, at up to 1,000 pixels on the long edge. HD & batch exports the original resolution and processes up to 50 images at once, still entirely on your device.