Google DeepMind Releases Genie 4 World Model With Audio

Google DeepMind released Genie 4, its latest interactive world model, in research preview on Tuesday, according to a report from TechCrunch. The system generates photorealistic, playable environments that respond to prompts typed mid-simulation and adds synchronized audio — music, ambient noise, and interactive dialogue — a capability absent from its predecessor, Genie 3, per TechCrunch.

What does Genie 4 add over Genie 3?

Genie 4 supports freeform mid-simulation edits, including changing weather, shifting camera angle, or inserting new people or objects, TechCrunch reported. The model also offers improved physics and more human-like characters compared with the prior version, according to the report.

What are the numbers behind the upgrade?

• Genie 4 can hold a consistent scene for up to two minutes — double the limit of Genie 3, TechCrunch reported.
• Google released Genie 3 as a research preview roughly a year before Genie 4's debut, per the report.

How does real-time steering work?

Steering refers to the ability to issue a text prompt while a simulation is running and have Genie 4 alter the scene without restarting it. Google DeepMind research director Tejas Kulkarni said during a briefing that the company expects "hybrid world models, where you have a combination of Gemini writing code and text, and Genie ingesting that in real time and simulating reality at different levels of abstraction."

How is Gemini integrated with Genie 4?

Gemini, Google's general AI model, now functions as a director that plans steps and checks frames while Genie 4 renders the visuals, according to TechCrunch. In a research experiment tied to Gemini Robotics, a Gemini 4 Argon model broke tasks into steps that Genie 4 simulated in real time — a method Google is testing as a way to validate robot behavior in simulation before deployment on physical hardware, the report said.

What are early testers using Genie 4 for?

Google's pitched commercial applications include gaming, interactive learning, and robotics, TechCrunch reported, citing a speculative example of a user entering a historical painting and conversing with its subjects. Early-access "trusted testers" have also flagged film prototyping as an emerging use. Google DeepMind engineer Diego Rivas said at a press briefing, "The real time aspect becomes most valuable in the areas where keeping a state of flow in the creativity space becomes incredibly useful — for example, in the ideation phase, in the storyboarding phase — anytime you're kind of in need of rapid ideation."

Who else is building world models?

The release lands amid consolidation in the sector, according to TechCrunch: AMD is acquiring Fei-Fei Li's World Labs, Amazon is in talks to buy Decart, and Rivian has acquired Third Dimension to build simulation tools for self-driving tests in-house. AI video company Runway released its own world model, GWM Worlds 2, with audio generation and dynamic response to user input, a couple of months before Genie 4's debut, the report said.