Google’s Genie 4 world model adds audio, more realistic worlds, and real-time steering
TechCrunch Rebecca Bellan ● Covered by 3 sources
Google’s new Genie 4 builds more realistic, steerable fake worlds and adds sound. It’s a big step for interactive AI, but the model still drifts and isn’t product-ready.
Based on reporting by TechCrunch, Rebecca Bellan — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Google DeepMind has put Genie 4 into research preview, and the pitch is simple enough: make AI worlds look better, behave better, and react while you’re still inside them. The new model is Google’s most photorealistic world generator yet, and it can take new instructions mid-simulation instead of waiting for a fresh prompt.
That matters because Genie 4 isn’t just drawing prettier scenery. It can add audio on the fly too — music, ambient noise, and dialogue that matches what’s happening in the simulation. Users can also change the weather, shift the camera angle, or introduce new people and objects while the scene is running. Google says the model can keep a consistent world going for up to two minutes, which is twice the limit of Genie 3.
There’s also a tighter link with Gemini, Google’s AI model. In this setup, Gemini acts more like a director: planning steps, checking frames, and helping Genie produce the visuals. In one research experiment tied to Gemini Robotics, a Gemini 4 Argon model broke tasks into steps that Genie 4 simulated in real time. Google’s Tejas Kulkarni described the future as a mix of code, text, and simulation, all feeding into each other.
The obvious uses are the usual trio: gaming, interactive learning, and robotics. But Google’s own testers are already imagining something messier and more interesting — film prototyping, where the value comes from fast iteration during storyboarding and ideation. That sits alongside Veo, Google’s other AI video model, which makes synchronized clips from text, image, or video prompts, and the company says Gemini Omni could eventually turn those prototypes into creative assets.
Still, Google keeps saying this is early. Genie 4 has gaps: details and objects can drift over longer runs, audio accuracy isn’t solved, and in-world text can still be hard to read. For now it’s available to trusted testers, with rollout planned for Google AI Ultra subscribers in Project Genie and for early access partners in the Gemini API and Google AI Studio.
My take — AI-written commentary, not fact-checked reporting
This is the part of AI that actually feels new: not another clip generator, but a system that can keep a world alive while you poke at it. The catch is the same old one — the demo looks magical right up until the model loses a shoe, a sign, or the plot. Everyone wants a “world model,” but nobody wants the bill for reality maintenance.
Read more about this at: TechCrunch