Google DeepMind has launched Genie 4, a new world model currently available in research preview. The model generates interactive, photorealistic environments and produces real-time audio, including music and ambient sounds. Users can modify scenes mid-simulation by typing prompts to adjust elements like weather, camera angles, or objects.
Genie 4 maintains scene consistency for up to two minutes, doubling the duration of its predecessor, Genie 3. The system features improved physics and more realistic human characters. Google has integrated its Gemini AI model to act as a director, planning tasks and verifying frames while Genie 4 handles visual generation.
The technology is being explored for applications in gaming, interactive education, and robotics simulation. Testers are also evaluating its potential for film prototyping and rapid storyboarding. Despite these advancements, the team notes that the model still faces challenges with object drift over time and the legibility of in-world text.
The release occurs amid consolidation in the world model sector. Recent industry activity includes AMD acquiring World Labs, Amazon entering talks to purchase Decart, and Rivian acquiring Third Dimension to enhance its internal simulation capabilities for self-driving vehicle testing.
