FWIW... Today's AI excels at finding patterns, but as some journalists have put it, it's closer to an advanced auto-correct than real intelligence. Since it's seen millions of images of people, & dogs & cats etc. [particularly cats], the text to image generators can do pretty much whatever you want with those common subjects, creating photo-realistic images if that's what you want. But ask for something less ordinary, and the results are more like graphics or artwork rather than photos, e.g., the Bing image creator & Firefly both came up with some Very cute pictures of a mouse paddling a canoe, but it was nowhere near photo-realistic. The AI image generators also have trouble with things like hands, and don't understand 3D space, &/or how parts of something fits together. You see the same sort of thing using AI to select parts of an image or photo.
Meta's Segment Anything works just about the same as Adobe's AI selection tools in Photoshop, & the more limited masking tools in Luminar Neo... they can get confused when there's not enough difference on both sides of an edge, if there's something extra [P/Shop often has problems with people wearing glasses if they're not looking at the camera], and/or with complicated shapes like some houses and buildings. [I tried Segment Anything with a photo of quite a few people outside of a complicated set of buildings, and it had no trouble at all selecting all the people, but the buildings were a complete mess.] In Photoshop & Luminar Neo you roughly paint over any parts it missed to add them to the selection or mask, & similarly paint over the stuff you don't want in the selection. [Something Meta left out of Segment Anything as it's just an online demo (though you can download & install it.)] And in Photoshop there are added, optional steps, e.g., tracing over the edge with another, related tool, which tells it to look much closer for stuff like hair or fur. PaintShop Pro's AI selection tool isn't nearly so advanced -- anything more than a simple portrait will probably require lots of old fashioned, hands-on work editing that selection. It does however have selection tools similar to Franzis Cutout, where you tell the software where the edge is generally, e.g. loosely tracing the edge with a wide brush, and the software looks for an abrupt color change within that area to find the edge.
I think it's great if you can use Segment Anything to help with editing. When the edges aren't perfect it would probably be a easier to clean those up in an image editing app than it would be to make the selection from scratch.
FWIW, restoring old photos from prints or negatives -- what I use P/Shop for most of the time nowadays -- I'm always using the AI focused Luminar Neo, the AI-based Topaz Denoise, and Adobe's AI neural filters. When I make selections it's usually the sky, & on these very old photos selecting the color range normally works better than the AI. Luminar Neo brings out a Lot of detail, sharpens, and often manages to make these grainy photos pop a bit, while the neural filters sometimes do a remarkable job with faces, and on some of the B&W photos, creating a colored version can be kinda cool.
