Two Google names are in the news right now. One is a real, announced model called Genie 4. The other, Gemini 4 Neon, is a name someone found in Google's code. I will separate what Google has actually said from what is rumor, explain what Genie 4 does in plain words, and give you a free prompt so you are ready when you get access.
- Genie 4 is Google DeepMind’s new world model, announced Oct 27. It builds an interactive scene you can steer by typing, and it now makes its own sound, including characters who talk back.
- Access is still a research preview for trusted partners. Google AI Ultra subscribers are next via Project Genie, with no date.
- Gemini 4 Neon is unconfirmed. A reference to it showed up in Gemini’s web code on Nov 16. Google has said nothing.
- Try it below: a free prompt that plans a scene and the edits you can type while it runs.
1. What Google announced: Genie 4
Announced Oct 27, 2026A quick note on timing. Genie 4 was announced on Oct 27, so it is not brand new. I am covering it now because the Gemini 4 family has kept rolling out since, and a new name, Neon, just appeared next to it. If you missed the Genie 4 news, this is the version with the context.
Genie 4 comes from Google DeepMind and it is what Google calls a world model. You type a description of a place, and the model generates it as a scene you can move around in, frame by frame, as you go. It is not a pre-rendered video. The scene reacts to what you do while it runs, which is why people compare it to a game that nobody built.
The big upgrade over Genie 3 is sound. Google says Genie 4 is its first real-time world model that makes audio natively. That covers ambient noise, music and dialogue, so a character you meet can answer you in the moment. TechCrunch also reports that a session can now run up to two minutes, double what the earlier model allowed.
The second upgrade is steering. While the scene is running you can type a plain sentence to change it, such as changing the weather, adding an object or bringing in another person. The keyboard handles movement and the camera. Under the hood, Google says Gemini Omni helps keep the world looking like the real one, and Google claims the model simulates real physics such as collisions and airflow.
2. Who can actually use Genie 4
Research previewRight now Genie 4 is a research preview for trusted testers and partners. Google says it is coming to Google AI Ultra subscribers inside Project Genie, the web app where the earlier version already lives. Google also says it will reach early access partners through the Gemini API and Google AI Studio. No date and no price have been given for any of that.
There is a waitlist for Project Genie users. Until you get in, a paid plan will not give you Genie 4, because Project Genie still runs the older model. I would hold off on upgrading just for this.
3. The robotics side most people skipped
Why Google caresThe most interesting demo is not a pretty landscape. Google showed a Gemini 4 Argon model breaking a task into steps and feeding those steps to Genie 4 as text while the scene rendered. In the demos a humanoid robot from Apptronik plays checkers, and in another a robot arm handles a pipette.
The idea is simple. A robot is expensive to test, and a real lab is slow. If a world model can show what a plan would look like in a believable physical scene, you can try ideas in simulation first. Be careful, though. Google describes these as simulations tied to its robotics research, and does not claim results on a real robot.
4. What we know about Gemini 4 Neon
UnconfirmedHere is the whole story on Neon. On Nov 16, the TestingCatalog account reported a reference to Gemini 4 Neon in Gemini’s web code, and said the model was not available yet. That is the only hard fact. Google has not announced anything, and I could not find a price, a date or an API name from an official source.
Everything past that is a guess, and the guessers say so. One guess is that Neon is a new Flash Lite model, meaning the cheapest and lightest tier. Another comes from an older report: in May, 9to5Google, citing code in an Android app, said Neon was a codename for a plan called Google AI Ultra Lite, sitting between the $20 Pro tier and the $250 Ultra tier. A third reading is that it is just an internal label. Treat all three as unproven.
For context, the Gemini 4 family has rolled out in tiers. Per one secondary report, Argon was announced on Sep 30, Carbon opened to paying users on Nov 10, and Hydrogen, the fast model, reached everyone on Nov 12 at $1.50 per million input tokens and $7.50 per million output tokens. If Neon is a cheap tier, it would fit a pattern of Google filling the range from top to bottom.
5. The limits Google admits
Read before you plan around itGoogle calls Genie 4 an early, experimental model and says so plainly. Details and objects can drift over time, especially after several complex edits or a lot of movement. Readable text, spatial accuracy and long-term consistency are listed as active research problems. Google also did not publish numbers to back up its demos, so the clips you have seen are curated.
That matters for creators. A two minute limit and visual drift mean this is a tool for short ideas, mood pieces and rough concept scenes, not for finished footage you can rely on yet.
6. My take: who it helps, the catch, the verdict
My opinionI have spent years cutting video, so my first reaction is about speed. Imagine testing a location, a mood and a camera move in a minute before you shoot or animate anything. That is who this helps: concept artists, game designers, directors and short-form creators who need to see an idea, not read about it.
The catch is access and control. You cannot use it yet, you cannot export clean footage from a two minute session, and a world that changes its mind halfway through is hard to build a project on. Add the missing price, and I would not plan a workflow around Genie 4 today.
On Neon, my honest advice is to wait. A leaked name in a web page is a reason to watch Google’s next announcement, not a reason to change your plan. My verdict: Genie 4 is the real story, a strong research preview worth following, and Neon is a rumor until Google confirms it.
7. Free prompt: plan a Genie-style scene before you get access
Unlocks in a few seconds- Wait for the timer, then click Show the prompt.
- Click Copy prompt and open a new chat in Claude.
- Replace the parts in [square brackets] with your idea and your goal.
- Send it. Save the scene prompt and the five edit prompts. Use the last part in any AI video tool today.
I want to be ready for AI "world model" tools like Google's Genie 4, where I type a scene and then steer it with short text edits while it runs. My idea: [one line, e.g. "a rainy night market where a street vendor chats with the player"] My goal: [e.g. "a 20-second moment for a YouTube Short" or "a game level idea" or "a training scene"] Please give me: 1. A scene prompt (3 to 5 sentences) that describes the place, the lighting, the sounds and one character who can talk back. 2. Five mid-run edit prompts I can type while the scene is running, each under 12 words (change the weather, add an object, change the mood, move the camera, add a second character). 3. A "drift checklist": 6 things to check at the 30-second and 60-second marks, because world models can lose details or object permanence over time. 4. What to do if the scene falls apart: a short recovery prompt and when to restart instead. 5. A version of the scene prompt I could paste into a normal AI video tool, if I do not have access to a world model yet. Use plain words and keep the whole answer under one page.
| Edit | What it changes | Check at 30 s |
|---|---|---|
| Make it start raining harder | Weather | Do the lights still match? |
| Add a street cat near the stall | New object | Is the cat still there? |
| Warm the colors, slow the music | Mood | Did the voices stay the same? |
| Move the camera behind the vendor | Camera | Is the stall still in place? |
| A second customer walks up | New person | Do both characters keep their look? |
8. What to watch next
Coming up- Project Genie access: Google says Genie 4 is coming to Ultra subscribers, with no date yet.
- A Neon announcement: any official word on whether it is a model, a plan or neither.
- Real numbers: resolution, price and session limits for Genie 4, which Google has not shared.
FAQ
What is Google Genie 4?
Genie 4 is a Google DeepMind world model. You describe a place and it generates an interactive scene in real time, now with built-in sound, and you can change it by typing short instructions while it runs.
Can I use Genie 4 now?
Not yet for most people. It is a research preview for trusted testers and partners. Google says Google AI Ultra subscribers will get it in Project Genie, and early access partners will get it through the Gemini API and AI Studio, with no date given.
What is Gemini 4 Neon?
Nobody outside Google knows. A reference to it appeared in Gemini’s web code on Nov 16. It might be a cheaper Flash Lite model or a subscription plan, but Google has confirmed nothing.
How long can a Genie 4 session run?
TechCrunch reports up to two minutes, double the previous model. Google’s own pages do not give a figure.
Want Claude to help you make the videos?

While world models catch up, Claude can already edit video for you. In my course Learn Video Editing with Claude Code you edit with Claude, with no Premiere and no code.
Get the course





