Remove Background
Remove an image background with an in-browser machine-learning model.
Remove an image background with an in-browser machine-learning model.
Removing a background used to mean tracing around a subject by hand. This page uses a background-removal model that runs inside your browser to separate the main subject from what is behind it, then saves the result as a PNG with a transparent background. It works best on pictures with a clear subject, such as a person, a product on a table or a pet.
When you click the button, the page first loads the model, which is the slowest part, and then processes your image. The status line under the button tells you what is happening, starting with the model loading and ending with a Done message once the file has been saved. Expect the first run to take longer than later ones because the model has to be fetched before use.
The tool makes a single automatic cut and offers no brush for touching up edges. For many everyday jobs that is enough, and the transparent PNG can be placed on a colored backdrop, in a slide or in a design file. For fine hair, glass or tricky shadows, plan to refine the cutout in an editor afterward.
Marketplace and shop listings look cleaner when the item sits on a plain background. Cut the product out, then place the transparent PNG on white or a brand color in your layout program.
A portrait without its original backdrop can be dropped onto a solid color or a gradient to match a team page or social profile. Start with a photo where your head and shoulders are clearly separated from the wall behind.
A cutout image looks more deliberate on a presentation slide than a rectangle with a messy background. Remove the background, save the PNG, and layer it over text or a colored shape.
When you are combining several subjects into one scene, each needs a transparent background first. Process each photo separately, then stack the resulting PNG files in whatever design tool you prefer.
Your image is processed in your browser by the background-removal model and is not sent to a server for editing. The model itself is loaded onto the page the first time you run the tool.
The page loads a machine-learning model that has been trained to tell the main subject of a picture from the area around it. It estimates which pixels belong to the subject and makes the rest transparent, then saves the result as a PNG. All of that runs inside your browser.
The model must be loaded into the page before it can process anything, and that takes longer than the cutout itself. The status text says when the model is loading. Later runs on the same page generally start working sooner.
Yes. The file is saved as a PNG, which supports transparency, so the removed area is see-through rather than white. If you open it in a viewer that shows a checkerboard or a dark backdrop, that is the transparency, not a flaw.
Pictures with one obvious subject, even lighting and good contrast against the background. People, products and pets on plain surfaces usually do well. Busy scenes, low light, motion blur and subjects that match their surroundings in color are harder.
No. The tool makes one automatic pass and saves the result. If the edges are rough or part of the subject is missing, open the PNG in an image editor and refine it, or try again with a sharper or better-lit photo.
No. The image is handled by the model running in your own browser, not by a remote editing service. The only thing the page needs to bring in is the model files used to do the processing.