Google DeepMind releases Genie 4 world model with audio and improved physics

Google DeepMind has announced the release of Genie 4, the latest iteration of its AI world model. The new model introduces significant improvements over its predecessor, Genie 3, including native audio generation, more realistic physics simulations, and enhanced consistency in generated environments. According to Google, these updates are designed to support applications ranging from robotics training to creative prototyping.

Enhanced Realism and Physics

Genie 4 leverages Gemini 3.8 Flash as a guiding agent to break down user goals and evaluate each output frame in real time. This integration helps maintain context as users move through varying environments. DeepMind states that the model has reached a new level of realism in simulating physics, particularly regarding aerodynamics and collisions. Additionally, human figures within the simulations are expected to exhibit more natural movement and responses to environmental changes, reducing the "uncanny" appearance often associated with AI-generated characters.

The model also supports midstream parameter changes. Users can alter weather conditions, add elements, or adjust camera angles using natural language prompts, with Gemini helping to steer the simulation accordingly. This process is similar to how reasoning models generate internal prompts to refine outputs.

Native Audio and Longer Consistency

A key feature of Genie 4 is the addition of native audio generation. This includes ambient background sounds and the ability to specify audio elements within prompts. DeepMind promises that the model can create interactive dialogue that adapts as the simulation changes. This marks the first time a DeepMind world model has included native audio capabilities.

In terms of duration, Genie 4 can create experiences that are approximately two minutes long, an increase from the one-minute limit of Genie 3. However, DeepMind notes that worlds still remain consistent for only a couple of minutes before the model "forgets" the initial context. The company warns that as simulations run longer, details and object permanence can shift over time.

Access and Pricing

Genie 4 is currently launching as a "research preview" to a small group of trusted testers. Google plans to roll out access to developers via the Gemini API and AI Studio, though specific pricing for these services has not yet been announced. For consumers, access will be phased in, starting with Google’s most expensive AI Ultra plans, which begin at $100 per month. Google may eventually expand access to AI Pro and AI Plus plans.

The consumer experience will be handled through Project Genie, which imposes certain limitations on usage. A waitlist is currently open for users interested in early access. DeepMind emphasizes that world models remain an active area of research, and Genie 3 never expanded beyond the limited Project Genie web experience.

Why it matters

The release of Genie 4 highlights the growing importance of world models in the development of artificial general intelligence (AGI). By enabling the creation of infinite, interactive training environments for physical robots and rapid prototyping for creative fields, these models offer new tools for both industrial and artistic applications. The addition of audio and improved physics brings these simulations closer to real-world fidelity, potentially accelerating their adoption in gaming, film pre-visualization, and robotics.

---

Sources

• arstechnica.com: DeepMind's Genie 4 world model adds audio and more realistic people to simulations - Ars Technica
• theverge.com: Google’s next Genie AI model generates more realistic experiences | The Verge

This article was drafted with AI assistance and checked against the sources above. Company claims are reported as claims. Cover image is AI-generated.