Google Deepmind has unveiled Genie 4, a real-time world model designed to simulate interactive environments. Unlike traditional video generators, the system calculates scenes frame by frame, allowing users to navigate through them using keyboard inputs and text prompts. The model is currently available only to a select group of testers.
A key feature of Genie 4 is its ability to generate sound directly, including background audio and character dialogue. The system leverages the Gemini Omni video model for visual realism, while other Gemini models handle simulation planning and consistency. For example, if a user changes weather conditions via text, the system adapts the scene accordingly.
Deepmind is also testing Gemini 4 Argon to integrate Genie with its robotics research. In a demonstration, the model simulated the Apptronik humanoid robot Apollo playing checkers by breaking the task into steps. The company notes that Genie 4 is an experimental model, and it currently faces challenges with object persistence and visual stability in long sessions.
To ensure transparency, all audio and video generated by the platform will include the SynthID watermark. Deepmind is continuing research into complex physical interactions and accurate text rendering. The company views the combination of Gemini and Genie as a step toward hybrid models that merge language-based reasoning with pixel-level simulation.
