Drop your video here or click to browse
MP4, WebM and MOV · your file never leaves this device
An on-device segmentation model separates the person from what's behind them. Blur the background, replace it with a colour, or make it green-screen for your editor.
Background removal used to mean hanging a green sheet behind you. A segmentation model does the same job by recognising the shape of a person in each frame and cutting everything else away — no fabric, no even lighting, no chroma key.
The model runs on your device. It is downloaded once from a public CDN and then does its work inside the tab, so your footage is never sent anywhere. That's a meaningful difference from the online background removers that require you to upload the clip and wait.
Blur is the most forgiving option, because a soft edge hides the small errors every segmentation model makes around hair and fingers. A solid colour is cleaner for talking-head clips over a busy room. Green screen is there so you can drop the export into a proper editor and key in whatever background you like.
The model estimates a person-shaped mask thirty times a second, and fine detail like stray hair is where it struggles. Even lighting, a plain background and standing further from what's behind you all help. Blur hides most of it.
It handles a group reasonably well when everyone is facing the camera, but it is trained on the single-presenter case and gets less reliable with several people at different depths.
Every frame goes through the model before it is drawn and encoded. Choose 480p if it can't keep up, and trim the clip to just what you need first.