01
Change the world as it unfolds
Navigation is only the beginning. Google describes natural-language edits to the weather, camera, objects and characters during an ongoing simulation.
REAL-TIME STEERING RESEARCH PREVIEW / OCT 2026
Genie 4.
Worlds that
respond.
Google DeepMind’s new interactive world model brings simulated worlds to life with native audio and real-time control. Understand what’s real, what’s available, and what comes next.
Independent guide · Checked October 28, 2026 · Primary sources ↗
THE WORLD MODEL OBSERVATORY
WORLD / 004
IMAGINATION → INTERACTION
IMAGINATION → INTERACTION
A new perspective
on possibility.
++on possibility.
CONCEPT ILLUSTRATION · NOT MODEL OUTPUT
THE LATEST
Announced October 27. Trusted-tester preview now; broader access is coming.
Google’s announcementA DIFFERENT KIND OF GENERATION
Don’t just watch.
Step inside.
A video plays out a sequence. A world model responds to what you do next.
Genie 4 generates the visible environment as an interaction unfolds. A user can explore a scene, adjust it with language and observe how it changes. The useful distinction is the feedback loop: your next action can influence the next moment.
That makes it interesting for creative exploration, interactive concept development and research into simulated environments. It also makes evaluation more demanding. A convincing clip does not, by itself, establish reliable physics, a persistent 3D map or the ability to export a finished game.
DESCRIBEEXPLORERESHAPE
WHAT CHANGED
Four ways the world
feels more alive.
Capabilities described in Google DeepMind’s
October 27 research announcement.
02
Listen to the world around you
Ambient sound and responsive dialogue are generated alongside the visuals. Audio is a native part of the model, rather than a soundtrack added to a silent recording.
NATIVE AUDIO03
Look closer. See more life.
The announcement highlights more natural human movement and improved physical interactions. Treat the demonstrations as examples of a research system, rather than a guarantee for every scene.
RICHER SIMULATION04
Give an idea a sequence of steps
Gemini models help plan and steer events as the world develops. This combines scene generation with reasoning, opening a path toward more controllable simulations.
GEMINI + GENIESEE IT IN CONTEXT
Watch the
official reveal.
Google DeepMind’s own introduction is the clearest starting point. Look for the connection between actions, visual changes and sound.
This is an embedded demonstration, not a playable instance. The video belongs to Google DeepMind.
Open the film on YouTubeGOOGLE DEEPMIND · OFFICIAL DEMONSTRATION
WHERE THINGS STAND
Curious to try it?
Start at the source.
Genie 4 is a research preview. Google’s linked waitlist is the route to express interest in Project Genie access. There is no model running on this website.
Join Google’s official waitlistResearch previewStarted with trusted testers
Consumer accessComing to Google AI Ultra in Project Genie
Developer accessEarly-access partners in Gemini API & AI Studio
Dates & standalone pricingNot specified in the announcement
Availability checked October 28, 2026. Eligibility and rollout can change.GO A LITTLE FURTHER
A guide for your
next question.
01 / FIELD NOTES
How to access Genie 4
Find the official Genie 4 waitlist, understand the research preview and Google AI Ultra rollout, and check what is known about pricing and access.
Explore the guide →02 / FIELD NOTES
Genie 4 vs Genie 3
Compare Genie 4 and Genie 3 using their official announcements: native audio, real-time steering, model limits and what the published specifications do not say.
Explore the guide →03 / FIELD NOTES
A practical Genie 4 prompt guide
Write clearer Genie 4 scene briefs with an original worked example, viewpoint constraints and a repeatable method for testing one change at a time.
Explore the guide →04 / FIELD NOTES
Genie 4 prompt examples
Browse and copy 16 original Genie 4 scene ideas for exploration, continuity, audio and dialogue, with a clear observation goal for every example.
Explore the guide →05 / FIELD NOTES
Genie 4 native audio & dialogue
Understand Genie 4 native audio and dialogue, then use an original listening checklist to evaluate timing, sound sources, perspective and scene continuity.
Explore the guide →06 / FIELD NOTES
Genie 4 limitations & evaluation
Understand the stated Genie 4 limitations and use an original run-log template to evaluate continuity, spatial relationships, audio and physical interactions.
Explore the guide →A PRACTICAL STARTING POINT
Think like an explorer.
Test like a researcher.
Describe a scene you can evaluate
Choose a small setting, a clear viewpoint and one meaningful action. A quiet courtyard with a single moving object is easier to assess than a busy city with dozens of competing events.
Change one thing at a time
Keep the camera or route repeatable. Then alter the weather, an object or a character instruction. Compare the result with the previous state before adding another change.
Record what holds up—and what drifts
Revisit a landmark, listen for audio matching visible events and note whether objects keep their identity. Save the prompt and the observation separately; an instruction is not evidence that it succeeded.
Original editorial evaluation advice. These are suggested experiments, not tested performance claims or a Genie 4 API specification.
AN ORIGINAL SCENE BRIEF
“A sunlit greenhouse. First-person viewpoint. Walk slowly along the central path as rain begins on the glass roof. Keep the red watering can beside the door.”
A starting idea to adapt when you have access. Results are not guaranteed.KEEP EXPECTATIONS GROUNDED
A remarkable preview.
An evolving system.
Google lists object permanence, spatial accuracy, complex physical interactions, legible text, audio accuracy and longer-term consistency as ongoing research areas.
Use simulations to explore ideas
Assess a camera angle, a visual direction or a scenario. Do not treat generated events as a measurement of what would happen in the real world.
Separate a demo from a product promise
Published demonstrations show particular examples. They do not establish a universal session length, an uptime guarantee, a commercial license, or production game-engine compatibility.
Check the version and the date
Project Genie is the experience; Genie 4 is a model. Instructions written for an earlier model may not describe the new preview. Follow the current official access and help pages.
CLEAR ANSWERS
The questions
worth asking.
Access, pricing and what the model actually does—without the guesswork.
What is Genie 4?
Genie 4 is Google DeepMind’s interactive world model announced on October 27, 2026. It generates a responsive simulated environment with native audio and supports real-time navigation and text-based steering.
Can I use Genie 4 right now?
The official announcement describes a research preview for trusted testers. It says access is coming to Google AI Ultra subscribers in Project Genie and early-access developer partners. Joining the official waitlist does not guarantee access or a date.
Is Genie 4 free, and what does it cost?
A standalone Genie 4 price has not been provided in the announcement. Google AI Ultra is named as an upcoming consumer access route; an existing subscription price is not proof of a Genie 4-specific price or entitlement.
Is there a Genie 4 API or a model download?
Early-access partners in the Gemini API and Google AI Studio are mentioned, but that does not establish a generally available API, public model weights, or a downloadable application. Check Google’s documentation before planning an integration.
Can Genie 4 create a complete playable game?
An explorable generated scene is different from a shipped game. Persistent saves, authored game rules, multiplayer, a production asset pipeline and game-engine export should not be assumed unless separately documented.
Is this the official Genie 4 website?
No. Genie 4 Space is an independent editorial resource. Google DeepMind develops Genie 4. Our official-resource links take you to Google’s announcement or its linked waitlist; we do not collect waitlist applications or provide model access.
FOLLOW THE EVIDENCE
Built on primary sources.
Written and maintained by Genie 4 Space Editorial. Last reviewed October 28, 2026. We summarize published information and clearly label our own suggested experiments. We have not independently tested the restricted model.