Yeah. You train it to recognize an object/image, and then you can respond to that however you want, programmatically. It's really cool for things like tabletop games.

For the how: Apple has APIs to including lightweight ML models for that sort of thing. You have to train a model, but for something like a card game it's super easy. For physical objects it requires more prep just because you have to take photos instead of using existing artwork... but it's still relatively straightforward.