Google DeepMind announced Genie 4 on October 27, 2026, the newest version of its "world model" line of AI systems. Unlike a video generator, which produces a fixed clip, a world model produces a place: an interactive, real-time simulation a user can move through, steer with plain-language instructions, and change while it runs. Genie 4 adds sound generated natively alongside the picture — ambient noise, music, and dialogue that adapts to events in the scene.
The release is a research preview, open first to a limited group of trusted testers. Google says access will expand to Google AI Ultra subscribers through Project Genie, and to early-access partners through the Gemini API and Google AI Studio. A public waitlist is open, but Google has announced no date or price for general availability.
Background: From Video Clips to Simulated Worlds
The distinction between a video model and a world model is the core idea behind Genie 4. A video model takes a prompt and returns a finished clip. A world model returns an environment that keeps running: the camera moves, characters react, and the scene responds to new instructions. The model must therefore maintain a consistent sense of space, objects, physics, and cause and effect over time.
Google's Genie line has been iterating on this problem. Earlier versions produced interactive scenes but with shorter coherent runs, no native sound, and less lifelike characters. Genie 4 pushes on all three fronts: a scene now holds together for about two minutes, roughly double Genie 3's window, with improved physics — including aerodynamics and collisions — and more humanlike characters. Google describes it as its first real-time world model to generate audio natively, meaning sound is produced as part of the simulation rather than layered on afterward.
Under the hood, the system leans on the Gemini family. Gemini models plan the scene, keep it coherent as it evolves, and adapt user instructions to what is already on screen. One report notes that Genie 4 uses Gemini 3.8 Flash for this directing role. In a robotics experiment cited by Google, Gemini broke tasks into steps that Genie 4 then simulated in real time, pointing toward robots trained inside generated worlds before touching the physical one.
What Happened
Google DeepMind announced Genie 4 on October 27, 2026, framing it as its "most realistic and interactive world model yet." The headline capabilities, as described by Google and reported across the technology press, are:
- Real-time interactive simulation. Users launch a world from a text or image prompt and steer it while it runs, typing instructions to change weather, camera angles, or the contents of a scene.
- Native audio. Ambient sound, music, and interactive dialogue are generated together with the visuals. Characters converse and react to user input or to events in the world, in real time.
- More realistic physics and characters. The model simulates physical scenarios such as aerodynamics and collisions, with characters that move and respond to their surroundings more naturally than in previous versions.
- Longer coherence. Scenes stay consistent for up to about two minutes, up from about one minute for Genie 3, according to TechCrunch.
- Watermarking. All generated video and audio carries SynthID, Google's hidden watermark for AI-made content.
Access is deliberately narrow for now. Trusted testers across filmmaking, journalism, education, game development, and engineering are trying the model first. Google says enterprise partners are testing it to generate scenarios quickly, and game developers are using it to go from an idea to an explorable prototype before writing engine code.
What Is Still Missing
Genie 4 is early, and Google says so. Visual details and object permanence can drift in longer sessions, text rendering remains weak, and audio accuracy is imperfect. A DeepMind engineer told reporters it is "still early days." A two-minute coherent scene is a research milestone, not a replacement for a game engine or a film set.
The sector is consolidating around world models as a strategic capability — AMD recently acquired Fei-Fei Li's World Labs, and Rivian bought Third Dimension — while Runway had already shipped its own playable world model with audio, GWM Worlds 2, about two months earlier. Genie 4 is Google's answer in a race now clearly about interactive simulation, not just video generation.
Why It Matters for Founders and Investors
World models collapse the cost of prototyping anything that lives in a simulated space: a game level, a training scenario, a product demo, a robot's test environment. A startup pitching a robotics system or training platform will soon be expected to demo inside a generated world rather than on slides, and founders who learn to build on these models early will have a head start when the APIs open up.
For investors, Genie 4 signals where the next platform layer is forming. Video generation was the last wave; interactive simulation is the next, pulling together gaming, robotics, film pre-visualisation, and enterprise training.
Angeled.com connects founders with investors.
TechCrunch, October 28, 2026
The AI Insider, October 28, 2026
Industry Wired, October 28, 2026
Gadgets 360, October 28, 2026
Read Next
OpenAI Releases Codex Security as Open-Source CLI and SDK
Databricks Unveils Always-On Genie, an AI Agent That Stops Waiting to Be Asked
DeepMind Launches AlphaGenome 2, a DNA Model That Reads Genomes Cell by Cell
ElevenLabs Unveils 13 New Products and Upgrades at Its New York Summit
Google Launches Gemini Decision 1, a Model That Answers Instead of Writing
Runway Open-Sources Praxis-1, Its Robot Model That Learned by Watching Video



