I just tested the Gemini 3 preview as well, and its capabilities are honestly surprising. As an experiment I asked it to recreate a small slice of Zelda , nothing fancy, just a mock interface and a very rough combat scene. It managed to put together a pretty convincing UI using only SVG, and even wired up some simple interactions.

It’s obviously nowhere near a real game, but the fact that it can structure and render something that coherent from a single prompt is kind of wild. Curious to see how far this generation can actually go once the tooling matures.