Remove Image Background, Free
Cut a person out of a photo for free, no sign-up, no watermark, no limit on how many photos. Everything runs in your browser, the image never leaves your device.
Click to choose a photo, drag and drop it here, or paste with Ctrl+V
Works best with one person, clearly in frame. Select multiple photos to batch process.
Your cutout will appear here once processed.
Common questions
Is this really free, with no sign-up?
Yes. There's no account, no watermark, and no limit on how many photos you process. The tool runs entirely in your browser using a free, open-source segmentation model, there's no server cost per image for us to recoup.
Does it work on product photos, not just people?
It's tuned specifically for people, portraits and selfies work best. Results on objects or products will be less reliable since the underlying model wasn't trained for that.
Is my photo uploaded to a server?
No. Processing happens entirely on your device after a small model downloads once per browser session. Your photo is never sent anywhere.
How it works
The heavy lifting is done by MODNet, an open-source portrait-matting network, run through ONNX Runtime Web (a WebAssembly build of the ONNX inference engine) inside your browser. The first time you click the button, the page downloads the model and runtime (about 18 MB combined, cached for the session), then runs inference on your photo locally. Unlike a plain segmentation model, MODNet predicts a real alpha matte, a per-pixel opacity value, rather than a blunt "person or background" guess, so hair and soft edges come out closer to true transparency.
Even a real alpha matte has some noise at its edges, so the tool still does a light cleanup pass: a small box blur smooths resizing artifacts, then the alpha curve gets a gentle smoothstep remap so background haze near zero snaps fully transparent and confident subject pixels snap fully opaque, while the soft transition in between (hair, blur, motion) is preserved rather than flattened. The cleaned result is applied per-pixel through canvas getImageData, and exports as PNG (to keep transparency) or composited over a solid color you pick.
This is also why results differ from paid cloud services: those run models hundreds of times larger on GPUs, so they win on hair and busy backgrounds. The tradeoff here is the opposite one, and it's deliberate: your photo never leaves your machine. That's verifiable, not a promise: open DevTools' Network tab and process an image. You'll see the model and runtime files download, and no request carrying your photo, ever. The model is trained on people, so it works best when a person is the main subject; products and pets get mixed results.