Google Genie 4: Release Date, Access and What the AI World Model Does

Google DeepMind just dropped Genie 4, and it's the closest thing yet to typing a game into existence. Announced Tuesday, October 27, 2026, it builds explorable 3D worlds in real time, now with its own sound, and lets you rewrite the scene mid-run with a text prompt. You can't play it yet unless you're on Google's tester list. Here's what it does, who gets it next, and why game devs are already poking at it.

Genie 4 release date and who can use it

• Announced: Tuesday, October 27, 2026, as a research preview.
• Who has it now: Google's "trusted testers" only.
• Next up: Google AI Ultra subscribers through Project Genie, plus early-access partners on the Gemini API and Google AI Studio. Google says "coming soon"; TechCrunch reports a rollout in the coming weeks.
• Waitlist: DeepMind's announcement post links a sign-up form.
• Free tier or standalone app: none announced.

What's new in Genie 4

• Native audio: ambient sound, music and interactive dialogue generated alongside the visuals. Genie 3 was silent.
• Edit the world live: type a prompt mid-session to change the weather, move the camera, or drop in people and objects.
• Keyboard controls: you move and look around with standard keys, like a game.
• Better physics and people: collisions, aerodynamics, and characters that react to the world around them.
• Longer runs: TechCrunch reports scenes stay consistent for up to two minutes, double Genie 3.
• Gemini under the hood: a Gemini model plans steps and checks frames while Genie draws the world.
• Watermarked: every generated video and audio clip carries Google's SynthID watermark.

Google is upfront about the limits. Details and object permanence drift in longer sessions, after lots of edits, or while you move around. Audio accuracy, legible in-world text and long-term consistency are still open research problems. A Google lead on the project told TechCrunch it's "still early days for Genie 4."

What it means for game devs and creators

DeepMind says game developers are already testing Genie 4 to go from an idea to an explorable prototype in seconds, trying out art styles, materials and mechanics in plain English before writing any engine code. Testers also see uses in film storyboarding. Google is pitching gaming next to robotics and education, so this isn't a gaming-only tool.

It also lands the same week Epic changed its Unreal Engine EULA to allow AI training on engine and MetaHuman outputs. The big platforms are all heading the same direction.

The 2nd Shift take

Genie 4 is a prototyping toy, not a game engine, and that's fine. Two minutes of drifting dream-world isn't replacing a level designer, and there's no save file or game loop here. But "describe a vibe, walk around in it, hear it" is a real shortcut for pitching ideas, and the audio jump is bigger than it sounds. The worry is the flood that follows. Steam is already fighting AI clones, and tools like this make cheap knockoffs even cheaper. Use it to sketch. Build the actual game yourself.

Want to know where AI fits in your own creative setup? Take the CRAFTR Quiz, then come argue about world models with us in the 2nd Shift Discord.

Keep reading: Epic will allow AI training on Unreal Engine and MetaHuman outputs and AI clone fears are pushing indie devs to cancel demos.

Sources: Google DeepMind, TechCrunch, Game Developer.