World models: how close are we to something usable in a real product?

I'm a dad of two (8 and 10) building a voice-first learning game for kids 6-12. Think Carmen Sandiego, but the kid is inside the adventure, talking to characters and solving the plot as they learn. Today I'm using 2D Rive animations driven by LLM reactions. Kids engage, but the ceiling is low. What I actually want is a real-time rendered character and world that the agent can direct moment to moment. So I've been tracking Genie 3, Odyssey, World Labs, and the avatar side (Runway, Anam). My wor