Remove Background from Photos with People
Remove the background from portraits, headshots, and photos of people — AI runs entirely in your browser. No upload, no server, no account required.
Drop PHOTO files here
or click to browse · batch supported
Choose filesHow browser-based AI background removal works
Everything happens locally in your browser using MODNet — a lightweight AI model optimised specifically for separating people from their backgrounds. The model (~6 MB) is downloaded once on your first visit and then cached permanently in your browser. After that, every image is processed entirely on your device without any network requests. Your photos never leave your machine at any stage.
MODNet (Modulate Objectness Deep Network) is a CNN architecture designed for real-time portrait matting. It analyses the image to identify the foreground subject and constructs a high-resolution alpha mask that captures fine edge detail — hair, transparent fabrics, blurred backgrounds. The mask is applied to the original image and the result is exported as a PNG with full transparency where the background was.
The output is always a PNG file because PNG is the only web image format with universal support for full alpha-channel transparency. You can open the result in any design tool — Figma, Photoshop, Canva, or even a basic image editor — and place it over any background. JPEG would fill the transparent areas with white; PNG keeps them genuinely transparent.
Frequently asked questions
This tool is optimised for photos containing people — portraits, headshots, full-body shots, and group photos. For product photos, animals, objects, or scenes without people, the AI model may produce inaccurate or incomplete cutouts. General-purpose background removal that handles any image type requires a larger, different model and is a planned future addition to this site.
Your photos are never uploaded anywhere. The AI model is downloaded once to your browser and then runs locally using WebAssembly. All processing happens on your device without any image data leaving it. You can verify this by opening your browser's DevTools Network panel while using the tool — you will see zero image data transmitted after the model finishes loading.
On your first visit, the browser needs to download and initialise the AI model (~6 MB). Subsequent images on the same visit, and all future visits once the model is cached, are processed much faster — typically under one second per image. The wait is a one-time cost, not a per-image overhead.
PNG is the only widely supported format that preserves full alpha-channel transparency. The background-removed result is a foreground subject with a transparent cutout where the background was — and that requires a format that supports alpha. JPEG does not support transparency at all. WebP does, but PNG has universal compatibility across all design tools, operating systems, and browsers without exception.