What is background removal?
Background removal — also called image matting or cutout — separates the subject of a photo from everything behind it and writes the result into the alpha channel of a PNG. Because the transparency is real pixel data rather than a white box, the cutout drops straight into product listings, ad creatives, thumbnails and design mockups.
This tool runs an open-source neural network entirely inside your browser. Nothing is uploaded, so there is no queue, no account, no file size limit imposed by a server, and no copy of your product photos sitting on someone else's disk.
How it works
- Drop up to 20 images into the box above, or click to browse. PNG, JPG and WebP are all supported.
- The first run downloads the AI model once. Your browser caches it, so later visits start instantly.
- Each image is processed one at a time on your own device and previewed on a checkerboard so you can see exactly what became transparent.
- Download results individually as PNG, or grab the whole batch at once as a ZIP archive.
Which AI engine should I pick?
Fast runs in every browser and is tuned for people and pets. It is the smallest download and the quickest option on a CPU.
General handles products, objects, animals and people. It needs WebGPU (a recent Chrome, Edge or Safari) because the model only ships in half precision, and it is roughly four times larger to download.
You can switch engines at any time. Existing results are cleared when you do, because a mask produced by one model says nothing about what the other one would produce.
Getting clean edges
- Give the subject some contrast with its background. A dark product on a dark surface is hard for any model.
- Fill more of the frame with the subject. A small object lost in a wide shot is easy to miss.
- Very fine detail — flyaway hair, fur, wire mesh — keeps softer edges, and the result may not be as clean as dedicated desktop editing software can produce. These models run entirely inside your browser, so they work within the memory and speed a browser can offer.