The curator's reviewDescribe any theme - croissants, thunderstorms, 1970s kitchens - and AI generates a complete chess set in that style, then you play a game with it.
GenChess is a small, perfect idea from Google Labs (late 2024): describe anything - breakfast foods, deep-sea creatures, municipal infrastructure - and an image model designs a full chess set in that theme: king, queen, bishop, knight, rook, pawn, each a coherent sculpture of your concept. Then you play an actual game with them, against a themed set the machine designs to oppose yours.
The design brief it solves per piece is genuinely hard: every piece must read as its chess role AND as your theme AND as part of one family. Watching it decide that the croissant is obviously the bishop - the curve, the deference, the diagonal energy - is watching a model do visual analogy under constraint, which is the actual frontier skill hiding under the whimsy.
The delight compounds when the opposing set arrives. Ask for breakfast and it might field condiments against you; the board becomes an argument between categories. And then, crucially, there is a real chess engine underneath - you can lose a legitimate game of chess to a squadron of hot sauces, an experience that builds character.
It is disposable in the best way: five minutes, one great screenshot, a genuine smile. But the croissant-as-bishop decision is the whole modern image-model story in miniature - the machine did not retrieve that answer; it composed it. My knights were pancakes. They moved in an L, syrup notwithstanding. Everything was correct and nothing was normal.
