On October 27, 2026, Google DeepMind, Google's AI research lab, announced Genie 4. It is an AI that builds a 3D world in real time from a text prompt.
People can walk through the world and change it with plain words, and its characters can now talk back. For now only invited testers can use it, and Google says it will come later to subscribers of its top AI plan (source).
What does this mean, and what comes next? What should people like this page's readers do now: people who make games, videos or lessons, people who use AI apps and agents at work, and people who just pay for an AI subscription?
In plain words: Google DeepMind — Google's artificial intelligence research lab; Genie 4 — Google's AI that builds explorable 3D worlds; World model — AI that simulates a world you can explore; Google AI Ultra — Google's most expensive paid AI subscription plan; Alphabet — the company that owns Google; Unity — company making software used to build games
What to do
Don't upgrade for Genie 4: write one two-minute test scene by October 30 and wait for a date.
Join one waitlist today, prepare one two-minute test scene, and don't pay extra for a preview.
Who should do what
Game, video and lesson makers: By October 30, write one short scene and three changes to test, and use it only for drafts.
People paying for an AI plan: Today, join the waitlist if you have the top plan; if not, don't upgrade until a date is posted.
People using AI helpers at work: Watch for tests of AI rehearsing tasks in made-up worlds; change nothing in your daily work yet.
What not to do: Don't buy any «Genie 4 invite» or upgrade your plan before Google posts a date.
You can only watch Genie 4 today; you can't use it. Only invited testers get in, and Google says its top plan comes «in the coming weeks». A world holds for about two minutes.
The costly mistake is paying early. That top plan costs roughly $200 a month, and there is no date or country list yet. Paid «invites» offered today are not real.
Prepare now, so you are ready when it opens. Makers of games, videos or lessons: by Friday, October 30, write down one short scene and three changes you would test. If you already pay for the top plan, join the waitlist today; if not, keep your current plan and your normal tools for finished work.
This is what most AIs advise. Each AI explains it in its own words below.
What the world's leading AIs think
They agree
Most answers: it's a preview for rough drafts, not finished work.
They split
Z.ai reports a gaming-stock drop today; Claude says there's been no repeat.
Not obvious
Making a world is solved for now; keeping it steady past two minutes is not.
What to watch
If Google gives a date and price for Ultra access by December, the conclusion holds.
The answers agree on the main point. Genie 4 is a preview, not a product. Only invited testers can use it. A scene holds together for about two minutes, double the last version, and then objects start to drift. Most see it as a tool for sketching games, film scenes and lessons, not for shipping them.
They also agree on money. Access is promised first to subscribers of Google's top plan, in the coming weeks, so nobody should upgrade just for this. Several point out that this plan costs about $200 a month, which makes waiting the cheaper choice.
They split on the market. One answer says game company shares fell 4% to 9% today. Another says Alphabet rose about 2% and the January fall in Unity's shares was not repeated. The January figures differ too: 21%, 24% and «more than 20%». Treat any stock number here as unproven.
They also split on how far this goes. Most stay with games and film. One idea goes further: a made-up world as a rehearsal room where an AI helper practises a task before doing it for real. Google shows this only as research, and a good rehearsal does not prove it is safe at work.
They split: Z.ai reports a gaming-stock drop today; Claude says there's been no repeat. What settles the dispute: Tonight's Alphabet earnings call and the next days' game-stock prices will show who is right.
No clear split: AIs from both countries gave the same advice, though China's side leaned more on stocks and the earnings call.
Who wins, who loses: Small creators gain fast drafts; engine and asset sellers face new pressure.
The final answer, written by an AI
Next to each AI: its place today and its score out of 10, given by other AIs acting as judges. Tap an answer to read it.
1KimiChina8.0Google's talking AI worlds hold together for just two minutes
Conclusion first: this is a lab preview, not a product you can touch today — and its most telling number is two, the minutes a generated world stays solid before details start to slip, double the old version but still toy-sized. Only invited testers can walk these talking, sounding worlds right now; everyone else gets demo videos.
For people who make games, films or lessons, the rough-draft era just got closer: film testers stepped into a live scene and reshot it several times while it ran, and photographers turned one flat archive photo into a walkable moment from history.
It still cannot replace a real game engine — objects and details drift — which is why Unity's boss calls it a helper, not a killer, even though Unity shares fell more than 20% the day Google's earlier Genie reached paying users. Rivals are racing too: Runway already has a similar sound-capable model, and AMD is buying a world-model startup.
Act today, Wednesday: if you pay for Google's $200-a-month top AI plan, join the Project Genie waitlist now, because Google says Genie 4 reaches those subscribers in the coming weeks.
If you are on the cheaper plan or none, do not upgrade just for this — instead watch Alphabet's earnings tonight after the market closes. And if you build games or videos for a living, sketch one small scene this week that you could pre-shoot inside a talking world, so you are ready the day the doors open.
Its own forecast: By December 31, 2026, at least one major game studio or film studio announces a paid partnership with Google to use Genie 4, going beyond today's tester program — 35% chance by Dec 31, 2026
2ChatGPTUSA8.0The next AI workplace may be a rehearsal room.
The big change is that AI could become somewhere you go, rather than something you watch or chat with. Imagine walking through a history lesson, testing a room design, or practicing a difficult conversation with someone who answers back.
But don't mistake the demonstrations for finished products: Genie 4's scenes currently last about two minutes, and objects can change unexpectedly (techcrunch.com). That makes it more promising for trying ideas than for replacing complete games or reliable training.
For game makers, filmmakers and teachers, the immediate opportunity is testing ideas before spending weeks building them; Google already describes game prototypes and film planning as early uses (deepmind.google). For people using AI helpers at work, the bigger possibility is rehearsal: an assistant could practice navigating changing situations before acting in the real world.
Google has demonstrated a system where one AI plans and checks actions while Genie creates the surroundings, although this is still research, not proof that workplace tasks can be safely automated (deepmind.google).
Check, AI subscribers, on October 28: if you want to experiment, join Google's official Genie 4 waitlist; if not, keep your current subscription and wait for independent tests (theverge.com).
Creators should write down one small scene they want to test, including what must stay consistent when they change it. When access arrives, judge whether it saves actual work, not whether its first few seconds look spectacular. My expectation is that short creative previews will become useful sooner than dependable, long-running virtual worlds.
Its own forecast: By November 27, 2026, Google DeepMind will publish at least one new official Genie 4 demonstration explicitly showing an interactive simulation lasting longer than two continuous minutes. — 28% chance by Nov 27, 2026
3ClaudeUSA7.6The last Genie sank Unity 24% in a day; this one hasn't
For most people, Genie 4 is a preview of the future and not yet a tool you can use. Only invited testers have it, and a waitlist is open (deepmind.google). It is a 'world model', meaning an AI that draws a living 3D scene frame by frame as you move. It now makes its own sound, music and talking characters (deepmind.google).
A scene stays consistent for up to two minutes, twice as long as Genie 3 (techcrunch.com). Google itself calls it early and experimental (deepmind.google).
Pictures still drift over long sessions, and audio accuracy, on-screen text and complex physical contact remain weak spots (investing.com). Everything it makes carries SynthID, a hidden watermark that marks it as AI-made (unite.ai).
Creators get the clearest use, but it is for sketching ideas, not finished work. A Google engineer said testers find it most useful for film prototypes and storyboards, which are rough plans of scenes, and he stressed it is early days (theaiinsider.tech). Game stock owners have a reason to watch closely.
When the earlier Genie came out in January, Unity closed down 24% in one day, its worst drop since 2022 (bloomberg.com). One report says Genie 4 has not set off a similar reaction (gpts24.com).
Unity also signed a partnership with Google this month to make games from text prompts (cryptobriefing.com). Alphabet shares rose more than 2% after the news (tradingkey.com). The company reports quarterly results after the market closes today, with analysts expecting about $128.2 billion in sales (cnbc.com).
Decide today, Wednesday, October 28, whether you need this. If you are a game, video or lesson maker who already pays for Google's top plan, Google AI Ultra, join the Project Genie waitlist now. Google says access comes 'in the coming weeks' (techcrunch.com). Plan to test it on rough drafts only. If you don't pay for Ultra, don't upgrade just for this.
There is no date yet, and Google has not said which countries will get it (pasqualepillitteri.it). People who use AI agents at work should note that developer access starts with selected partners through Google's developer tools (Gemini API and Google AI Studio), not with everyone (deepmind.google).
Its own forecast: Google officially makes Genie 4 (in Project Genie) available to Google AI Ultra subscribers in at least one country outside the United States, per an official Google or DeepMind post or help page — 20% chance by Dec 31, 2026
4Nemotron + MercuryUSA7.4Genie 4 holds a talking 3D world steady for 2 minutes — Ultra users may get it in weeks
For an everyday AI user, Genie 4 is a new kind of tool that turns a text description into a moving, sounding 3D scene you can walk through and change by speaking.
Right now only invited testers can try it; Google says its top-tier AI Ultra subscribers will get access "in the coming weeks" through Project Genie, and developers can start with the Gemini API and Google AI Studio. That means your current subscription doesn't change today, but a real 3D world builder could arrive before December.
The people whose next step shifts now are game makers, video producers, teachers, and robotics researchers — DeepMind has already put Genie 4 in their hands as trusted testers. If you build games or lessons, you can join the official waitlist and start thinking of scenes you'd want to prototype.
Investors should watch Alphabet's Q3 earnings today (analysts expect about $127.8 billion) and any mention of Genie on the call, since the last Genie announcement moved related stocks.
If you create games or lessons, by Friday October 30 sign up on the Genie 4 waitlist and write down one prompt you'd love to see become a playable scene; if you only pay for an AI subscription, don't upgrade yet — check Google's AI blog once a week and only switch plans when a page dated on or before November 27 confirms Genie 4 is live for Ultra users.
Its own forecast: Google DeepMind publishes a blog post showing a full game level created with Genie 4 in under five minutes — 40% chance by Nov 27, 2026
5Z.aiChina7.0Invited testers only — even Google's top-paying subscribers must wait
For regular people this is a preview, not a product: only invited testers can walk through Genie 4 worlds today, and Google has only promised later access to subscribers of its top AI plan and to developers (deepmind.google).
When it opens, the pitch is a game that builds itself while you play — you say 'make the bridge ice' and it happens, and the people in the world answer out loud. Treat it as a pricey toy at first, not something your evening plans depend on.
Game makers have the most at stake: worlds that appear from plain words threaten the slow, hand-built way games are made, and shares of game companies — Unity (whose software builds many games), Electronic Arts, Take-Two, Roblox — fell between 4% and 9% in early trading today (techcrunch.com).
Alphabet investors get the next episode tonight, October 28: the company reports its July-to-September results after Wall Street closes, with experts expecting about $128.0 billion in revenue (bloomberg.com). Teachers and video makers should watch for the developer opening — a world you can walk through and rewrite by voice is a lesson or a storyboard you never had to build.
Ignore every 'Genie 4 invite' for sale today, Wednesday, October 28. If you already pay for Google's top AI plan, watch the DeepMind blog and your Google app in the coming weeks and try Genie 4 the day it lands there; if not, don't buy a pricier plan for one gated preview — wait for the wider opening, then judge the worlds yourself. Waiting a month costs nothing; buying hype can cost real money.
Its own forecast: Unity Software (NYSE: U) closes at least 10% above its October 28, 2026 closing price on some trading day by November 27, 2026, per Google Finance daily closes — 55% chance by Nov 27, 2026
This is typing that turns into a place you can walk through. Genie 4 adds real sound, music and people who talk back in real time, and lets you change weather or camera while it runs. Worlds still drift if you stay long and they only hold together for about two minutes, so it is a preview, not a finished tool.
For folks who build games, videos or lessons, the job moves from laying every brick to directing what happens. Filmmakers are running several takes in seconds, teachers turning one old photo into a room students can enter, game teams sketching a playable draft before writing any engine code. That speed is why gaming stocks slid in January, with Unity down 21%, and why engines now face a new kind of prototype.
Check as a maker tonight, October 28, if you already pay for Ultra join the Project Genie waitlist; if not, keep your current editor and watch tester clips until Google posts a price and date.
Its own forecast: Variety, The Hollywood Reporter or Deadline reports a film or TV production used Genie 4 for pre-visualization — 55% chance by Nov 20, 2026
7GeminiUSA6.8Genie 4 holds worlds steady for two minutes before shapes melt
Computer programs can now create walkable 3D places with background noises and speaking people from a sentence, but they still forget where objects were after about two minutes. For everyday people, this means interactive computer movies and custom game scenes are turning into something anyone can steer without technical training, though they remain experimental toys rather than polished software.
This shift challenges game studios and online teachers to treat artificial intelligence as a live scene-builder instead of just a drawing tool. At the same time, companies building standard 3D game engines keep their edge because professional games need exact physical rules and lasting memory that these models still lack.
Check your project needs on October 28, 2026: if you create quick concept videos or prototype levels, join Google's testing waitlist right away; if you build finished games or commercial training tools, keep using established game engines until live simulations solve memory drift.
Its own forecast: Google will publicly announce an integrated export feature allowing developers to save Genie 3D environments into standard game engine file formats (such as USD or glTF). — 45% chance by Apr 30, 2027
8QwenChina6.5Google's new AI lets users walk through 3D scenes just by talking
Google's new Genie 4 model turns plain text into a 3D world you can walk through and change in real time. It keeps the scene stable for two full minutes, which is double the limit of last year's version, and now adds realistic sounds and talking characters. This means ordinary people can soon type a prompt and instantly step inside a custom video game or interactive movie without needing to code or animate anything.
Filmmakers and game builders are already using it as a rapid drafting tool to prototype scenes, adjust camera angles on the fly, and test visual ideas before writing engine code.
Teachers could soon turn a flat history photo into a populated, walkable simulation where students can chat with historical figures. For robotics companies, the system pairs with Google's logic models to safely test multi-step physical tasks in a simulated environment before trying them in the real world.
Check the Project Genie waitlist today, October 28, and join it if you are an AI fan; if not, just wait for the public release. The tool still struggles with keeping objects in place over long runs, so expect a few visual glitches while you play.
Its own forecast: Google will release a public update fixing Genie 4's object permanence drift for all Ultra subscribers — 30% chance by Dec 15, 2026
9WenxinChina6.4Genie 4 adds talking characters — game-building cost could drop to near zero
People who make games, videos or lessons should treat this as a preview of tools that let anyone type a sentence and walk inside the result — the gap between "having an idea" and "having a playable scene" just shrank dramatically.
What matters now isn't the technical details but the speed: if Genie 4 lets a teacher describe a rainforest and students walk through it while characters explain the food chain, entire lesson-prep budgets could be rethought within months.
Businesses that sell game engines, stock 3D assets, or hire small animation studios face the biggest shift — Unity and similar companies saw their shares drop sharply after the first Genie, and today's richer version with sound and dialogue deepens that pressure. Investors should watch Alphabet's earnings call today (October 28, 2026) for any mention of Genie 4 revenue plans; a clear rollout date would be the signal that matters.
If you pay for an AI subscription right now, wait for the official rollout announcement before assuming Genie 4 is included — Google has only said "later" for top-tier subscribers.
If you make creative content, start experimenting with today's publicly available AI video and 3D tools so you understand the workflow Genie 4 will eventually compress into one prompt; if you only use AI at work for text tasks, bookmark this story and check back in 30 days for concrete access details.
Its own forecast: Google announces a specific Genie 4 launch date for paying subscribers (not just a waitlist) — 55% chance by Nov 15, 2026
10LongCatChina5.8Genie 4 keeps a world alive for 2 minutes — twice Genie 3's limit
It means the AI can now build a small, changeable 3D world with sound and talking characters, but only for about two minutes at a time — so it is a sketching tool, not a finished product. For an ordinary person, nothing changes today: you cannot use it, and the only thing worth doing is joining the waitlist if you are curious.
It changes things for people who make games, videos or lessons. They can now test an idea in minutes instead of weeks, but they still need real tools to turn it into something people can buy or watch. People who use AI at work should watch, not switch — the model is too short and too shaky for real tasks yet.
Today, October 28, join the waitlist if you want to try it first; if you just pay for an AI subscription, wait until it is actually open to you before changing anything.
