Google DeepMind has launched Genie 4, a world model that generates interactive, real-time simulations with native audio. Google says the model brings a new level of realism to world simulation and is rolling out now as a research preview to a limited group of trusted testers.
Genie 4 is built on the Gemini family. Gemini Omni anchors each simulation in real-world appearance and behavior, while other Gemini models supply the reasoning that keeps a scene coherent and continuous as it runs. Launch reports describe Gemini 3.8 Flash working as an in-loop agent that plans steps and checks generated frames.
The headline addition is sound. Each world generates its own audio on the fly, including ambient noise, music and interactive dialogue from characters. That makes Genie 4 the lab's first real-time world model to produce audio natively; its predecessor, Genie 3, generated silent worlds.
Key facts
- 01Google DeepMind announced Genie 4, a world model for real-time interactive simulation, on Oct. 27, 2026.
- 02Genie 4 generates native audio, including ambient sound, music and interactive character dialogue.
- 03Users can change weather, move the camera or add characters mid-session with natural-language prompts.
- 04Trusted testers have access now; Google AI Ultra subscribers in Project Genie and Gemini API partners are next.
- 05Every video and audio clip Genie 4 produces carries an invisible SynthID watermark.
Users can also steer a running scene in plain language. Google's examples include moving the camera, changing the weather and adding characters, with the model adapting each edit to what is already on screen. DeepMind also claims better physics for aerodynamics, collisions and human motion, and calls simulated people the largest leap over the previous version.
Launch coverage puts Genie 4's scene consistency at up to about two minutes, roughly double Genie 3's limit. Google has not published resolution, frame rate, latency or benchmark figures for the model.
Access is staged. Trusted testers come first, followed by Google AI Ultra subscribers through Project Genie and early-access partners in the Gemini API and AI Studio. Google says those groups are coming soon, and a waitlist is open in Project Genie.
Every video and audio clip Genie 4 produces will carry a SynthID watermark. The marks are invisible and require a detection tool to check.
Google lists the model's current limits itself. Visual details and object permanence can drift in longer sessions or after complex edits, real-time audio accuracy is unresolved, and legible text inside generated worlds is not yet reliable. DeepMind describes Genie 4 as an early, experimental model.
Google frames world simulation as a step toward AGI. The company points to safe training environments for robots and AI agents, along with creative tools for filmmaking, journalism, gaming and manufacturing, as the main uses for the model.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
