Editorial illustration for Google's AI Worlds Gain Voice in New Update
Google's Genie 4 Adds Voice to AI-Generated Worlds
Google's AI Worlds Gain Voice in New Update
Google DeepMind's last world model, Genie 3, let people walk around an AI-generated scene for a few minutes before it fell apart. Genie 4, announced this week, adds something Genie 3 never had: a voice. The new system generates its own ambient sound and music, and populates its worlds with characters who can hold live conversations with the person exploring them.
The mechanics stay familiar. Users still move through scenes with keyboard controls and can drop in text prompts mid-exploration to change what's happening, with DeepMind saying the world stays coherent as those changes land. What's new is who's in it and what they can do. DeepMind paired the model with an Apptronik Apollo robot, simulating it inside a Genie 4 world to test how a physical machine might operate in a space that only exists as a generated scene.
That detail points to where DeepMind wants this technology to go next: not just worlds people can look at and walk through, but worlds robots can train in before they ever touch anything real. Here's what the company showed off to make that case.
Genie 4 is the first Genie model to generate its own sound, like background noise and music, with characters in the worlds also now able to hold live conversations.
Why this matters
Genie 2 and Genie 3 proved Google could generate a world you could walk through. Genie 4 proves it can populate that world with something that talks back, and that's a different kind of milestone. A simulation with voice and sound isn't just more immersive, it's more useful as training data. That checkers-playing humanoid robot demo is the tell: DeepMind isn't building Genie 4 as a toy for wandering virtual landscapes, it's building a sandbox where robots and agents can rehearse physical and social interaction before they ever touch the real world.
For researchers, that's a cheaper, faster way to generate edge cases than waiting on real-world robotics data. For founders building agentic or embodied AI products, it's worth watching whether Google opens this up beyond demos, because a world model that handles dialogue and physics in one system starts to look like infrastructure, not a research toy. The gap between "watch an AI world" and "negotiate with the things inside it" just got a lot smaller, and that's the part worth tracking closely.
Related Reading
Browse all Google news →
Research & Benchmarks Microsoft AI Chief: "If It Isn't Safe We Shouldn't Build It
Research & Benchmarks Google's FACTS benchmark shows 70% factuality ceiling across four tests
Research & Benchmarks Nvidia's DLSS 4.5 beta adds 6x Multi Frame Generation for RTX 50 GPUs
Related Topic Google to End Free Access to Gemini AI for Non-Subscribers
LLMs & Generative AI
Related Topic NVIDIA and Google Cloud let developers scale AI from prototype to production
Market Trends
From Archives Oracle Adds Google TPUs, Cerebras to AI Compute Options
From Archives Maya Research Releases Open-Weight Speech Model With 12 Languages and 20 Voices
From Archives OpenAI Reportedly Plans to Purchase Google AI Chips
Open Source
From Archives DeepMind Genie 4 Modernizes Pre-LLM Agents Using Gemini Omni
LLMs & Generative AI
Common Questions Answered
What new audio capabilities does Genie 4 add compared to Genie 3?
Genie 4 is the first Genie model to generate its own ambient sound and music within AI-generated worlds, a feature that Genie 3 completely lacked. Additionally, characters in Genie 4's worlds can now hold live conversations with users exploring the environment, making the experience significantly more interactive and immersive than its predecessor.
How do users interact with scenes in Genie 4's AI worlds?
Users navigate through Genie 4's AI-generated scenes using keyboard controls, similar to previous versions of the system. They can also drop in text prompts mid-exploration to modify or influence the world as they move through it, allowing for dynamic interaction with the environment.
Why is Genie 4's voice and sound generation significant beyond just entertainment?
A simulation with voice and sound creates more immersive and realistic training data that is more useful for developing AI agents and robots. DeepMind is positioning Genie 4 as a sandbox where robots and agents can rehearse and learn, rather than just a toy for exploring virtual landscapes, making it a valuable tool for advancing robotics and AI development.
What progression has Google DeepMind made from Genie 3 to Genie 4?
While Genie 3 allowed users to walk around AI-generated scenes for a few minutes before the simulation deteriorated, Genie 4 extends this capability with interactive voice and sound. The progression demonstrates Google's advancement from creating explorable worlds to creating responsive, conversational environments populated with characters that can engage with users.
Further Reading
- Papers with Code Benchmarks - Papers with Code
- Chatbot Arena Leaderboard - LMSYS
