Most tools solve the same hard problem
Every background remover estimates an alpha value per pixel. At edges each pixel mixes subject and background colour — three equations, seven unknowns — so there's no single right answer, only better or worse estimates [1]. That is why tools look identical on a sneaker and very different on curly hair.
Different models, different strengths
Public research models show the range: U²-Net finds salient objects at many scales [2], MODNet specialises in real-time portraits [3], IS-Net and BiRefNet target very fine, high-resolution detail [4][5], and Segment Anything segments whatever you click — as hard masks rather than soft alpha [6]. Commercial tools rarely say which model they run, so test on your own photos.
