Google DeepMind’s Genie 4 creates interactive worlds with native audio, real-time steering, improved physics and more realistic characters.
Google DeepMind has released Genie 4, a world model that creates interactive environments with video, sound and real-time control. Users can change a scene with natural-language prompts while the simulation is running. The model can alter weather, move the camera, add objects and generate dialogue that responds to events. Google is presenting Genie 4 as a research preview for trusted testers before a wider rollout.
Genie 4 adds sound to world simulation
Genie 4 generates audio directly inside its simulated environments. Its sounds can include music, wind, waves, ambient noise and interactive speech. That makes the model different from systems that create visual scenes first and add sound as a separate production step.
Google says the model is natively multimodal, meaning it can combine visual information, language, movement and audio in one experience. 
Google DeepMind’s Genie 4 announcement says characters can speak and react to user inputs in real time. However, Google also acknowledges that audio accuracy remains an active research problem.
Real-time prompts control the scene
The biggest change is the ability to steer a world while it unfolds. A user can type a request to change the weather, adjust the viewpoint or introduce a new person. Gemini helps interpret the instruction and adapt it to the existing scene, rather than treating every prompt as a completely new generation.
That control could make interactive storytelling more flexible. Filmmakers might test camera angles and scene ideas without building a full set. Game developers could explore mechanics before writing engine code, while educators could create environments that respond to questions and decisions. TechCrunch reported that Google researchers also see potential for rapid storyboarding and creative prototyping.
Gemini helps Genie maintain coherence
Google is connecting Genie 4 with Gemini models to handle planning and reasoning. Gemini can break a goal into smaller steps, watch the generated frames and check whether each step succeeded. Genie then renders the changing physical environment and responds as the plan develops.
The companies’ research team tested this approach with Gemini 4 Argon and Gemini Robotics. In one demonstration, the systems simulated a humanoid robot playing checkers by converting a user’s instruction into a sequence of actions. 
Google’s official overview of Genie 4 describes this combination as a path toward testing robot tasks in simulated environments.
The model targets robotics and creative work
World models attempt to predict how environments change when people or machines act within them. That makes them useful for more than entertainment. A robot could practise a task in a simulated warehouse, while developers could test how an agent handles unfamiliar objects or changing conditions.
Google also lists filmmaking, journalism, gaming and manufacturing as potential areas of interest. For journalists and educators, an interactive simulation could turn a photograph or historical setting into an explorable scene. For production teams, the model could offer a fast draft before expensive filming or animation begins.
These possibilities remain early-stage. As another example of the broader shift toward AI tools in everyday decision-making, readers can see Diaspora Digital Media’s report, 
“Bank of America Warns Customers Not to Share Banking Passwords With AI Assistants”.
Genie 4 still has important limits
Google says visual details and objects can drift during longer sessions. The problem becomes more likely after several complex edits or when users move through a scene. Physical interactions, readable text, spatial accuracy and long-term consistency also require further improvement.
The release is therefore not a finished consumer product. Google is using trusted testers to identify weaknesses before expanding access to Project Genie, Google AI Ultra subscribers, Gemini API partners and Google AI Studio users. The company says generated audio and video will include SynthID, an invisible watermark designed to help identify synthetic media.
“Genie 4 marks a new area of frontier research that merges real-time generation and photorealism.” — Google DeepMind
Ultimately, Genie 4 shows Google moving world models toward richer and more controllable simulations. Native sound and real-time steering make the experience closer to an interactive environment than a fixed video. The technology could support robotics, games, education and visual planning, but its reliability will determine how useful it becomes. For now, Genie 4 remains an ambitious research preview with significant work ahead.
