Genie 4

957 points · 548 comments on HN · read original →

Points and comments are a snapshot, not live.

DeepMind's Genie 4 generates real-time interactive 3D worlds with audio and speech.

Genie 4 is a multimodal world model that produces interactive simulations with photorealistic visuals, physics, ambient sounds, and responsive dialogue. It incorporates Gemini models for reasoning and coherence. Users can steer the world via keyboard and natural language prompts. The model is available as a research preview for trusted testers and will roll out to Google AI Ultra subscribers and API partners. Outputs are watermarked with SynthID. Limitations include visual drift over long sessions and imperfect physics and text rendering.

What commenters are saying

Commenters are deeply skeptical of Genie 4's utility. A top thread argues that world models will enable a new DRM regime where games are simulated on cloud servers and streamed, likening it to a failed Stadia model. Replies note Stadia failed due to latency and graphics degradation, not just business decisions, and that AI-generated games lack polish and are cost-prohibitive to stream. Others question the demo's interactivity, claiming inputs appear ignored. A few suggest the real use case is robotics simulation, not consumer gaming.