Remove a background — without uploading the photo
The AI model downloads to your browser once and then runs on your own device. No signup, no watermark, no credits, no limit.
Every other background remover uploads your photo to a server, hands you a low-resolution preview, and asks for money or a credit before you can download the real thing. That design exists because the model runs on their machine. This one downloads the model to yours — about 25MB, once, then cached — and after that your photos never leave your device and there is nothing to charge you for. Works best on people and portraits.
Original
Cutout
Questions people ask
Is my photo uploaded?
No. The model is downloaded to your browser and executed on your device with WebAssembly. Your photo is never transmitted — and the uploads counter in the top bar stays at 0 while you use it. This is the one claim every other background remover structurally cannot make: their model lives on their server, so your image has to go there.
Why is there a 25MB download?
It is about 25MB of model plus a roughly 13MB WebAssembly runtime the first time you use any on-device AI tool here — call it 38MB on a first visit, and 25MB if you have already used another one. Because the neural network runs on your machine instead of ours, and a neural network is a big file. That is the whole trade: one download your browser then caches, in exchange for no upload, no per-image credit and no account. We tell you the size before it starts rather than pulling it silently.
Is there a watermark, a credit system or a resolution limit?
None of them. What you download is the full-resolution cutout. There is no paid tier holding anything back, because there is no paid tier.
What does it work best on?
People. The model is trained for human matting, so portraits, headshots and full-body shots are its strength, including reasonably fine hair edges. Product shots and general objects will be less reliable — that is an honest limit of a model small enough to run in a browser.
Can I put a colour or a blur behind instead?
Yes. Choose transparent for a PNG cutout, a solid colour for passport-style or store-listing photos, or blur the original for a portrait-mode look. All three are composed on your device.
Does it work offline?
After the first run, yes. The model file is cached by your browser and by this site's service worker, so it keeps working with no connection.
How this works & why it’s private
This page runs MODNet, an open-source (Apache-2.0) image-matting network, through ONNX Runtime compiled to WebAssembly. Both the runtime and the model weights are served from this domain — not from a CDN — so no third party sees your photo, and no third party sees your IP address either. Your image is drawn to a <canvas>, converted to a tensor, run through the network on your CPU, and the resulting alpha matte is composited back onto the original pixels, all inside the tab. The only network request the tool makes is fetching the model file from this site, once; after that it is served from cache. Open your dev-tools Network tab and watch — your photo is never in a request body.