Published October 28, 2026. Last updated October 28, 2026.
Type a sentence. Get a world you can walk around in. Then type another sentence and change the weather while you’re standing in it.
That’s the pitch behind Genie 4, the new world model from Google DeepMind, announced October 27, 2026. Some of the early write-ups make it sound like finished magic. It isn’t. It’s an early research preview, and Google says so.
Here’s what Genie 4 actually does, what’s still missing, and who can try it.
Quick answer
What is Genie 4? A world model from Google DeepMind. It generates interactive, real-time virtual worlds from text or image prompts, with sound included.
What’s new? It adds native audio (music, ambient noise and dialogue), real-time steering with text prompts, and more realistic visuals. TechCrunch reports scenes stay consistent for up to two minutes, double Genie 3’s limit.
Can I use it today? Only if you’re a trusted tester. It’s “coming soon” to Google AI Ultra subscribers through Project Genie, and to early-access partners in the Gemini API and Google AI Studio. A waitlist is open.
Is it perfect? No. Google says details and objects can drift over time, especially in longer sessions.
What is a world model?
A world model is an AI that doesn’t just draw what a scene looks like. It tries to simulate how a world behaves, so things respond when you act.
Compare it with a video generator. A video model produces a clip, and you watch it. A world model produces a place, and you move through it. Press a key, the camera moves. Change the prompt, the world changes. That’s the difference between watching a film and playing a game.
Google’s own post doesn’t compare Genie 4 to specific video tools. That comparison is my explanation, not theirs.
What can Genie 4 do?
Based on Google DeepMind’s announcement:
- Native audio. Google calls it its first real-time world model to generate audio natively, including environmental ambience, music and dialogue that reacts to what happens.
- Real-time steering. You navigate and move the camera with standard keyboard controls. You can also type prompts mid-simulation to change the weather, move the camera, or add a person or object.
- Physics. Google says it simulates real-world physics such as aerodynamics and collisions, and that characters move naturally and respond to their surroundings.
- Multi-step events. Gemini 3.8 Flash acts as a guiding agent. It breaks goals into steps, checks the frames to confirm each step worked, and adjusts when the world changes or the user steps in.
Under the hood, Google says Gemini Omni keeps simulations anchored to real-world appearance and behavior, and other Gemini models handle reasoning and coherence. DeepMind’s researchers also talk about future “hybrid world models” that combine language-model reasoning with pixel-level simulation.
How is Genie 4 different from Genie 3?
Genie 3 arrived as a limited research preview in August 2025. According to Unite.ai, it ran at 24 frames per second and 720p. TechCrunch says Genie 4 is far more photorealistic and adds audio, freeform real-time edits, better physics and more human-like characters.
Google hasn’t published Genie 4’s resolution, frame rate or latency. So claims like “near-zero latency” aren’t backed by anything in the announcement. “Real time” is what Google says, and that’s the claim to hold on to.
Can robots really train inside Genie 4?
That’s one of Google’s stated goals, and there’s a demo. DeepMind combined Genie 4 with the Gemini 4 Argon model and Gemini Robotics to simulate an Apptronik Apollo humanoid playing checkers, with each task step shown as a text caption.
It’s a research experiment, not a product. It suggests a future where robots practice in simulation before they touch anything physical. If you want to know more about Argon, I covered it here: Google launches Gemini 4 Argon.
What else is Google pitching it for?
- Filmmaking: pre-visualization, where directors test multiple “takes” in a live simulation.
- Games: prototyping an explorable idea before writing any engine code.
- Education and history: populated worlds built from archival photos.
- Journalism and disaster visualization: turning a single photograph into a simulated environment, or visualizing events like floods and wildfires.
- Enterprise: fast scenario generation.
DeepMind engineer Diego Rivas told reporters that “The real time aspect becomes most valuable” for creative flow, especially in ideation and storyboarding. He also said some capabilities, like turning prototypes into finished creative assets with Gemini Omni, are “still a ways off.”
Who can use Genie 4 right now?
Almost nobody outside Google’s partners. It’s a research preview for trusted testers.
Google says the next steps are:
- Google AI Ultra subscribers, through Project Genie
- Early-access partners in the Gemini API and Google AI Studio
TechCrunch says the Ultra rollout is expected in the coming weeks. Google hasn’t named supported countries, pricing or a general release date, so don’t plan around any of those yet.
What are Genie 4’s limits?
Google is open about these:
- Visual details and object permanence can drift, especially in longer sessions, after many edits, or as you move through a scene.
- Complex physical interactions, real-time audio accuracy, readable text, spatial accuracy and long-term consistency are listed as ongoing research areas.
- The finished product experience isn’t clear yet.
That matters because some summaries call it “physics-accurate” with “persistent world states.” Google’s own wording is more careful. It simulates physics scenarios, but long-term consistency is still unsolved.
Is it safe?
Google says Genie 4 has strict safety guardrails like its other models, went through internal safety evaluations, and was released first to trusted testers. All generated video and audio carries SynthID watermarking, Google’s invisible marker for AI-made media.
Who else is building world models?
It’s a crowded race. TechCrunch notes that Runway released GWM Worlds 2, a playable model with audio, a couple of months ago. It also reports that AMD is acquiring Fei-Fei Li’s World Labs, Amazon is reportedly in talks to buy Decart, and Rivian acquired Third Dimension for self-driving simulation.
What does this mean for creators?
If you make videos, games or storyboards, this is worth watching, not worth rebuilding your workflow around yet. The most realistic near-term use is quick visual ideation: testing a scene, a camera angle or a mood before you commit time to making it.
Be careful with expectations. Two minutes of consistency and visible drift mean this isn’t a replacement for a game engine or a film crew. It’s a sketchpad that moves.
FAQ
What is Genie 4?
Genie 4 is Google DeepMind’s world model for real-time interactive simulation. It generates explorable worlds with native audio that you can steer with keyboard controls and text prompts.
When was Genie 4 announced?
October 27, 2026, as a research preview for trusted testers.
Is Genie 4 a video generator?
Not exactly. A video generator makes a clip you watch. Genie 4 makes a world you can move through and change as it runs.
How long can a Genie 4 simulation last?
TechCrunch reports scenes stay consistent for up to two minutes, double Genie 3’s limit. Google hasn’t published a maximum session length.
Can I try Genie 4?
Not yet unless you’re a trusted tester. It’s coming soon to Google AI Ultra subscribers through Project Genie, and a waitlist is open.
Does Genie 4 have sound?
Yes. Google says it generates music, ambient noise and interactive dialogue in real time.
How much does Genie 4 cost?
Google hasn’t announced pricing.
Bottom line
Genie 4 is a real step: interactive worlds, native audio and steering by text, all running live. It’s also an early preview with drift, short consistency windows and no public numbers on speed or resolution.
Watch it, join the waitlist if you’re curious, and wait for people outside Google to test it before you believe the biggest claims.