What the model is doing
The model looks at the picture and estimates which pixels belong to the foreground. Those pixels stay. The rest become transparent in the PNG you download. It is an estimate, so edges can be wrong when the subject and the background look similar.
What usually works, and what to inspect
- People, products, and objects with a visible outline tend to come out cleaner.
- Plain or simple backgrounds are easier than busy patterns.
- Fine hair, fur, glass, smoke, and motion blur need a close look in the preview.
- A logo made of simple shapes is often cleaner than a photograph of that logo.
Use the before-and-after slider on the result. If an edge is wrong, the honest next step is another source image with a clearer subject, not a promise that every file will be perfect.
Why the first run waits on a download
The model is not bundled into the first page load. It downloads when you start a removal, then stays in the browser cache for later images. Progress during that download is the real transfer progress, not a fake timer. Details of the free limits are on the free background remover page.