The model is small and inspectable
The pinned quantized weight is 6.6 MB, Apache-2.0 licensed and disclosed with its exact SHA-256.
Remove a portrait background — on your device
Drop in a portrait, let a small matting model isolate the person, then tune the edge and preview transparent, solid or blurred backgrounds. Your photo never leaves this device.
◉MODNet6.6 MBLOCAL
Drop a portrait hereCtrl/⌘ V or paste an imageJPEG, PNG or WebP · up to 50 MB
6.6 MB local modelDesigned for people
First use downloads the 6.6 MB model and local AI runtime. Supported browsers cache them for later offline use.
MODNet q8
Inspect the cutout
Inference, edge refinement and export stay in this browser. No request contains your photo.
Keep creating locally
Apache-2.0 · ONNX · WASM
A private portrait cutout, not a cloud upload
Most background removers send the photo to an API and hide useful downloads behind an account or quota. This tool ships a pinned, quantized MODNet model to the browser. The runtime sees local pixels, produces an alpha matte, and exports a high-resolution result on this device.
Edges remain under your control
Use edge and feather controls while checking transparency, solid colors or a blurred-photo background.
The result joins the workflow
Send the transparent PNG to Icon Studio, Crop, Compress or the pixel-level Image Compare Lab without selecting it again.
Local background removal questions
Does the image get uploaded?
No request contains your image. The first use downloads the model and ONNX runtime; inference and export then run locally.
Will it work for products or pets?
Sometimes, but this MODNet checkpoint is trained for portrait matting. The page does not claim general-object accuracy. People against a reasonably distinct background work best.
Why is the first run slower?
The browser must download and initialize about 6.6 MB of model weights plus the local inference runtime. The service worker caches used assets for later offline reuse.