Introducing Genie 4
Google releases Genie 4 world model with audio, photorealism, and Gemini-in-the-loop control.
“Audio modality is a net new modality”
Google DeepMind launched Genie 4, the latest iteration of its interactive world model series, adding audio generation, significantly improved photorealism via its Omni Flash foundation, and Gemini-in-the-loop control. The model represents a progression from Genie 3's real-time 3D environments toward more immersive, multimodal simulation — though the team emphasizes it remains in the research range. This matters as world models are increasingly positioned as infrastructure for training AI agents in rich simulated environments.