How to remove the background from a photo
- Drag a photo onto the upload area, or click to choose it. JPG, PNG and WebP all work.
- Pick an AI model: Any photo for products, cars, pets and objects (and people), or Portraits for photos of people.
- Click Remove background. The first time you use a model it downloads (about 5 or 7 MB) and you see a percentage. After that it starts straight away.
- Check the result next to the original. Choose Transparent, White, Black or any Colour for the new background.
- Click Download. A transparent background saves as PNG; a coloured one saves as JPG. The result is the same size as your photo.
To do another photo, click Start over.
Which model to choose
- Any photo (U²-Net) finds the main subject of the photo, whatever it is: a product for an online shop, a car, a pet, an object or a person. Start with this one.
- Portraits (MODNet) is trained only on people. It can keep finer detail around hair, but on anything that isn't a person it keeps parts of the background or loses the subject.
If one model gets your photo wrong, try the other on the same photo: it only takes a few seconds once both are downloaded.
Is my photo uploaded?
No. The AI models run inside your browser, so your photo stays on your device. What goes over the internet is the download of the models themselves, from Hugging Face (a site that hosts AI models), and of the engine that runs them, from the jsDelivr CDN. Your browser keeps them, so later uses don't download them again.
Limits
- Things touching the subject: a car door someone leans on, or a chair someone sits in, can stay in the result, sometimes faded. Photos where the subject stands clear of the background work best.
- Edges: hair, fur, glass and soft shadows are the hardest parts, and a little of the old background colour can stay around them. Check the result before you use it.
- Several subjects: the model keeps what looks like the main subject. Smaller things nearby, like a remote control next to a pet, may be kept too.
- One photo at a time.
To shrink the result for the web, use Image Compressor or Image Resizer.