Google DeepMind Unveils Genie 4 with Real-Time Audio and Steering

Genie 4 extends the capabilities of its predecessor by doubling the duration of consistent scene generation to two minutes while introducing improved physics and responsive characters. Central to this evolution is a tighter integration with Gemini, which now functions as a directorial layer, planning simulation steps and validating frames while Genie renders the visual output. This synergy, demonstrated through experiments with Gemini Robotics, suggests a path for testing complex robotic tasks within virtual environments before physical deployment.

Beyond technical benchmarks, the model targets creative workflows. While industry competitors like Runway and acquisitions by firms such as AMD and Amazon signal a consolidation in the world model sector, Google is positioning Genie 4 as a tool for rapid ideation. Testers are exploring its utility in film storyboarding, where the ability to maintain a state of creative flow is paramount. Despite these advancements, the team notes persistent challenges, including object drift during long simulations and difficulties with rendering legible in-world text.

The model is currently accessible to trusted testers, with a broader rollout scheduled in the coming weeks for Google AI Ultra subscribers within Project Genie and early access partners via the Gemini API and Google AI Studio.