What Google DeepMind's Genie 4 is
Genie 4 is what researchers call a world model, which is a step beyond the video generators people are used to. A tool like that makes a clip you watch. Genie 4 makes a place you can enter and change while it runs. The environments are photorealistic, they follow real-time physics, and they come with native audio rather than sound added afterwards. The characters inside them are responsive too, moving and reacting instead of standing as scenery. All of it is powered by Google's Gemini models working underneath.
Today we're introducing Genie 4 — our newest interactive world model from @GoogleDeepMind 🌐
How Genie 4 turns a prompt into a world
The part drawing the most attention is the control. You describe what you want in plain language, and Genie 4 generates the scene, then lets you keep steering it as it plays. That includes shaping what happens next and even moving the camera around, the kind of control you would normally expect from a game engine rather than an AI prompt box. Google's demos showed the range, from quiet snowy landscapes to crowded city streets, each one built and explorable in real time.
Why early testers are praising Genie 4
Google DeepMind set the tone in its own post, calling Genie 4 "our most realistic and interactive world model yet" and inviting people to explore dynamic environments and see "how the scene responds, sounds, and how the people in it move, react, and speak." The reaction from early testers matched the pitch. Justine Moore, an investor at a16z who got early access, singled out the leap in being able to create fully interactive, GTA-style worlds from one prompt, the kind of thing that used to take a studio and a budget.
When you can get access to Genie 4
For now the access is narrow. Google DeepMind is rolling Genie 4 out to a group of trusted users first, with wider access promised soon for Ultra subscribers, its top paid tier. So most people will be watching the demos rather than steering their own worlds for a little while yet. Those demos are worth watching anyway. One showed a crowned crane walking across fresh snow beneath a clear dusk sky, rendered in natural, documentary-style colour, and generated as a world you could step into rather than a clip you could only watch.
Standfirst: Google DeepMind has unveiled Genie 4, an interactive world model that builds playable, photorealistic worlds from a single text prompt. Early testers are calling it a major leap, with access rolling out now and Ultra subscribers next.