Google's Genie 4 world model gains audio, richer realism and real‑time streaming
deepmind google
| Source: TechCrunch · via Yahoo Tech | Original article
Google DeepMind unveiled Genie 4, a world model that now incorporates real‑time audio and generates more realistic simulations.
Google DeepMind has rolled out Genie 4, the latest iteration of its AI‑driven world model, in a research‑preview limited to trusted testers. Building on the Genie 3 platform that sparked a dip in gaming‑sector stocks last year, Genie 4 expands the simulation envelope with native, real‑time audio generation, richer physics, and more lifelike human avatars. The model can now sustain interactive scenes for up to two minutes—double the duration of its predecessor—while allowing users to inject text prompts mid‑simulation to reshape environments on the fly.
The upgrade is powered by Gemini 3.8 Flash, which preserves contextual continuity across the extended run time. Audio output spans music, ambient sounds and interactive dialogue, creating photorealistic worlds that respond instantly to user input. However, DeepMind notes that consistency remains limited to a few minutes, indicating that longer, seamless experiences are still a work in progress.
Genie 4 matters because world models are becoming a central focus for AI labs seeking to blur the line between virtual and physical environments. Enhanced realism and real‑time interactivity open pathways for more immersive gaming, advanced training simulators, and novel content‑creation tools. The ability to modify simulations on the spot also hints at future workflows where designers and developers iterate within a live, AI‑generated sandbox.
Watch for broader access beyond the current tester pool, potential integration into Google’s consumer or enterprise products, and how rivals respond with their own high‑fidelity world models. Continued improvements in temporal consistency and longer simulation windows will be key indicators of whether Genie 4 can move from research preview to commercial reality.
Sources
Back to AIPULSEN