Type a scene, step inside it, and change the weather while you walk around. That’s the pitch behind Genie 4.
Google DeepMind announced Genie 4 on Tuesday, describing it as its new world model for real-time interactive simulation. Unlike a video generator that hands you a finished clip, a world model keeps generating the scene as you move through it with standard keyboard controls, and responds when you steer it with text.
Now with sound
The biggest change is audio. DeepMind says Genie 4 is its first real-time world model to generate audio natively. Every world carries its own ambient sound, from wind in the trees to waves on a beach, and characters can hold interactive dialogue that reacts to your inputs or to events in the scene.
DeepMind also calls out a jump in realism, especially for people. Characters are meant to move naturally, respond to their surroundings and be steered in real time. The model is also designed to simulate physical scenarios such as aerodynamics and collisions.
Gemini in the loop
Genie 4 leans on the Gemini family. According to DeepMind, Gemini Omni anchors simulations in real-world appearance and behaviour, while other Gemini models handle the reasoning needed to keep a scene coherent. When you type a change like “introduce a new character”, Gemini adapts the prompt to fit the scene.
For more complex tasks, Gemini 3.8 Flash acts as a guiding agent: it watches the scene, breaks a goal into steps and checks incoming frames to confirm each step worked. DeepMind showed Gemini 4 Argon and Genie 4 working together to simulate a humanoid robot playing checkers.
Who can use it
Not many people yet. Genie 4 is rolling out as a research preview to trusted testers. Google says it is coming soon to Google AI Ultra subscribers in Project Genie and to early access partners in the Gemini API and Google AI Studio, with a waitlist open now.
Testers so far include people in film, journalism, education, gaming and engineering. Filmmakers are using it as a “draft mode” for pre-visualisation, game developers to prototype ideas before writing engine code, and photojournalists to turn a single archival photo into an explorable scene.
DeepMind is upfront about the limits. Visual details and object permanence can drift in longer sessions, and complex physics, audio accuracy, legible text and long-term consistency are still being worked on. All video and audio from Genie 4 carries SynthID, Google’s invisible watermark for AI-generated content.
Why it matters
World models are moving from research demos towards real tools. For creative agencies, game studios and training teams, the ability to sketch an interactive scene from a sentence could cut early concept work from days to minutes. For robotics, safe simulated worlds are a cheaper way to train machines before they touch the real world. The built-in watermarking is also a reminder that brands should plan for clear labelling of AI-made content.
Sources: Google DeepMind, Google. Written by the xRapid Media LLP team.