Google DeepMind ships Genie 4, its first world model with native audio
Genie 4 generates ambient sound, music and character dialogue alongside the simulated scene itself, and is rolling out to Gemini API early access partners.
Google DeepMind announced Genie 4 on 27 October, a world model that generates explorable, interactive environments in real time from a text or image prompt. The headline change from Genie 3: it’s DeepMind’s first world model to generate audio natively, producing ambient sound, music and character dialogue alongside the visuals as the scene unfolds.
You can steer a running scene with natural language, moving the camera, changing the weather, or introducing a new character, while a Gemini model handles planning to keep the world coherent. DeepMind says physics and character movement are both improved over Genie 3, though it’s upfront that object permanence can still drift the longer a simulation runs.
Genie 4 is rolling out now as a research preview to trusted testers only. DeepMind says it’s coming “in the coming weeks” to Google AI Ultra subscribers through Project Genie, and to early access partners of the Gemini API and Google AI Studio. No pricing, release date or API documentation yet.
Why it matters: this is the first sign that world models are heading toward the Gemini API rather than staying a research demo. If DeepMind follows its usual pattern, expect Argon- and Nano Banana-style general availability to follow the early access phase within weeks, not months.
The caveat: everything here is a research preview with no public API. There’s nothing to build against yet, and DeepMind’s own post flags scene drift as an open problem.