This tool uses IMG.LY's open-source background removal library, which runs an IS-Net segmentation model with ONNX Runtime and WebAssembly inside your browser. The model looks at the whole photo and estimates, for every pixel, how likely it is to belong to the main subject. That estimate becomes the alpha channel of the result: subject pixels stay opaque, background pixels become transparent, and edge pixels can be partly transparent, which helps with soft edges such as hair.
There are no settings to tune. The model decides what the subject is.