I tried it on a bee image. The bee was not a person, so it did not work.

I tried a few cartoon images. Realistic styles (most of Avatar: The Last Airbender, most of Ghost in the Shell) work well. Semi-deformed styles (Avatar gag faces, Samurai Jack, The Simpsons) usually didn't work or resulted in artifacting. Non-human cartoons (Spongebob, Adventure Time) never worked.

Some photographs which had low contrast between subject and background were not as cleanly cut out. It was good enough that I wouldbe able to import the mask as a starting point and clean up the edges manually.

Photographs in which faces were obscured where handled surprisingly well!

Would love to see this as a native feature of an image editor.

Yeah, make a (PS/other image editors) plugin that shove it into the API and return the mask that you can correct when needed, that seems the way to go.

From their about page: "Our AI is trained to detect persons as foreground and everything else as background. That's why it only works if there is at least one person in the image."

Think they're referring to one of the top 5 comments on any Show HN post being about how it doesn't work in mobile Safari or Firefox etc.

I tried it with a photo of a crowd of people and it chose around 20% of the people spread randomly around the photo and removed the others.