Google’s Genie 4 world model adds audio, more realistic worlds, and real-time steering
News Source
•Tue, 27 Oct 2026 16:30:57 +0000
📰 What Happened
Google DeepMind released Genie 4 on Tuesday as a research preview. It is a world model: an AI system that builds playable, moving scenes instead of just video clips. Users can type new instructions while a scene runs, such as changing the weather or moving the camera.
The model also makes sound that fits each world, including music, background noise and spoken lines. Scenes now stay consistent for up to two minutes, twice as long as Genie 3. Google says the physics and characters look more real, and its Gemini model now helps plan and check each frame.
🔍 The Backstory
World models are a newer goal than chatbots or image tools. Instead of guessing the next word or pixel, they try to copy how a place behaves, so a person or a robot can move through it and see what happens. Google first showed Genie as a research project in 2024.
The field is crowded and getting smaller. Runway released its second world model this year. AMD is buying World Labs, Amazon is in talks to buy Decart, and Rivian bought Third Dimension for self-driving tests. Big players are buying these skills rather than building them alone.
🎯 Why It Matters
World models could train robots and cars in fake worlds before they touch the real one. They also raise hard questions about realistic fake video that anyone can steer.
Google DeepMind released Genie 4 on Tuesday as a research preview. It is a world model: an AI system that builds playable, moving scenes instead of just video clips. Users can type new instructions while a scene runs, such as changing the weather or moving the camera.
The model also makes sound that fits each world, including music, background noise and spoken lines. Scenes now stay consistent for up to two minutes, twice as long as Genie 3. Google says the physics and characters look more real, and its Gemini model now helps plan and check each frame.
World models are a newer goal than chatbots or image tools. Instead of guessing the next word or pixel, they try to copy how a place behaves, so a person or a robot can move through it and see what happens. Google first showed Genie as a research project in 2024.
The field is crowded and getting smaller. Runway released its second world model this year. AMD is buying World Labs, Amazon is in talks to buy Decart, and Rivian bought Third Dimension for self-driving tests. Big players are buying these skills rather than building them alone.
World models could train robots and cars in fake worlds before they touch the real one. They also raise hard questions about realistic fake video that anyone can steer.