How to Remove an Image Background Free (and Get Clean Edges)

Removing a background used to mean tracing around a subject by hand with a selection tool. Today a neural network does the tracing. The model looks at your photo and decides, pixel by pixel, how likely each one is to belong to the main subject. That map of probabilities is called a mask.

The tool then applies the mask to your picture. Pixels that belong to the subject stay. Pixels that belong to the background become transparent, which software stores in a hidden extra channel called alpha. The model our remover uses is called ISNet, and the result is built at your photo's full resolution, up to 4096 pixels on the long edge.

The format you save in decides whether the transparency survives.

• If you plan to put the cutout on your own background in another app, export a transparent PNG or WebP instead of flattening it with a color.
• A JPEG will always have a solid background. If a tool exports JPEG for a transparent cutout, the transparent area turns white or black.

The model looks for a clear, dominant subject. The more obvious that subject is, the cleaner the cutout.

• Works well: a sharp subject that contrasts with the background, good light, and one main object or person.
• Harder: hair and fur, because single strands are thinner than a pixel and blend into the background.
• Harder: glass, water and anything see-through, where the background shows through the subject.
• Harder: a subject that is close in color and texture to what is behind it, like a white cup on a white table.
• Harder: cluttered scenes with several similar objects, and very dark or motion-blurred photos.

Here is the whole process.

• Open the Background Remover and add one photo, or many. JPG, PNG and WebP all work.
• Wait for the first photo. This is when the AI model downloads, about 88 MB, once per browser. After that it is cached, and every later photo starts straight away.
• Choose what goes behind the subject: transparent, a solid color, or the original background blurred.
• Adjust edge softness. Use 0 for products and logos, and a touch more for portraits with hair.
• Download the result. More than one photo comes as a ZIP file.

Most rough results can be improved without any editing skill.

• Soften the edge a step or two when you see a faint halo of the old background. It usually comes from compression artifacts at the original edge, not from the model.
• Start from the best photo you have. A larger, sharper original gives the model more to work with.
• When you shoot a new photo, put the subject in front of a plain backdrop and light it evenly. That helps more than any setting.
• Place the cutout on a background with similar lighting and color. Even a perfect cutout looks fake over a background that does not match.
• Check the edges at 100% zoom before you use the image. Small problems show up there first.

Common jobs include product photos for an online shop, profile pictures, presentation graphics, thumbnails, stickers and collages. If you sell online, check the image rules of the marketplace you use, because many have their own requirements for backgrounds and sizes.

Many background removers upload your photo to a server, apply a limit, then ask you to pay for the full resolution. Ours does the work in your browser, using WebAssembly, so the photo stays on your device.

The trade-off is that your browser has to download the model once, about 88 MB. If you are on mobile data, it is worth waiting for Wi-Fi. After the first time it is cached, so later visits are quick.