Background Remover
Cut people out of photos locally, with no credits and no uploads.
Drop a photo of a person here, or pick one with the button. Everything runs in this tab: your files and inputs never leave your device.
This tool runs MODNet, an open portrait matting model, inside your browser. The weights are a one time download of about 6.3 MB from this site, and your browser keeps them afterwards, so later visits start straight from the cache and work offline. Nothing is uploaded: your files and inputs never leave your device.
MODNet is a portrait matting model, so it is at its best on photos of people and gets noticeably vaguer on products, pets, and busy objects. The model always sees a copy of the photo scaled to about 512 pixels on its shortest side, and the matte is scaled back up and applied at full resolution, so the file you download keeps the size it came in at.
What it does
Runs MODNet, an open portrait matting model, inside your browser tab to separate a person from the background of a photo. The model predicts a soft alpha matte rather than a hard outline, so hair and fuzzy edges survive instead of turning into a jagged silhouette. You get the cutout as a transparent PNG, or composited onto white or any hex color as a JPEG. MODNet was trained on portraits, so it is strong on people and noticeably weaker on products, pets, and busy objects.
How to use it
Drop a photo onto the panel or pick one with the file button, then press Load model to fetch the 6.3 MB of weights once. Press Remove background and the cutout appears next to the original on a checkerboard so you can see exactly what the alpha channel is doing. Choose the background you want, leave feathering on unless the edge looks too soft, then download the result.
Why this one
The best known background remover charges per image and needs your photo uploaded to its servers before it will show you anything above thumbnail size. This one downloads an open portrait matting model once, runs it in the tab you already have open, and your files and inputs never leave your device. The honest trade is quality: on a person against a normal background it holds its own, and on a product shot or a complicated object a paid service will usually beat it.
FAQ
- Why do product shots come out worse than photos of people?
- MODNet is a portrait matting model. It was trained to find a human subject and predict the soft alpha around hair and clothing, so that is what it looks for in every image you give it. A bottle, a chair, or a pet has none of the cues it learned, so the matte gets vague and the edges wander. For non human subjects, the manual selection in a normal image editor is often faster than fixing what this returns.
- What image size can it handle?
- Anything up to 4096 pixels on the longest side. Larger photos are scaled down to that limit first, and the panel tells you when it does. The model itself always sees a smaller copy, around 512 pixels on the shortest side, which is what it was trained at. The matte is then scaled back up and applied to the full size photo, so the file you download keeps its original resolution.
- Are my photos uploaded anywhere?
- No. The model weights are downloaded from this site to your browser once, and every pixel of your photo is read, matted, and re-encoded inside the tab. Your files and inputs never leave your device, and once the weights are cached the tool keeps working with the network off.