Genie 4 builds a world you can walk into, and it talks back

The big story: Genie 4 makes worlds you can steer in real time

On October 27, 2026, Google DeepMind announced Genie 4, which it calls "our new world model" (Google DeepMind, Oct 27, 2026).

A world model is a different thing from a video generator. A video tool gives you a finished clip. Genie 4 gives you a scene you move around in with standard keyboard controls, and you change it with a text prompt while it runs: switch the weather, move the camera, or bring in a new person or object. DeepMind says Gemini adapts your prompt to fit the surrounding scene and keeps the world moving between your inputs (Google DeepMind).

We wrote a full explainer on it: Genie 4: Google's AI that builds a world you can walk into.

What is new in Genie 4

DeepMind highlights four upgrades (Google DeepMind, Oct 27, 2026):

• Real-time steering. Text prompts reshape the world as it plays, and Gemini monitors the scene to keep it coherent.
• More realistic physics and people. DeepMind says it simulates things like aerodynamics and collisions, and that the biggest leap is in how people move and react.
• Native sound. It is DeepMind's first real-time world model to generate audio itself, from wind and waves to characters who hold conversations and react to what you do.
• Multi-step events. Gemini 3.8 Flash acts as a guide in the loop. It breaks a goal into steps, watches the frames, and checks that each step worked. DeepMind also showed Gemini 4 Argon planning steps for a simulated Apptronik Apollo humanoid playing checkers.

Who can use it, and when

Not most of us, yet. Genie 4 is a research preview for trusted testers. DeepMind says it is coming soon to Google AI Ultra subscribers in Project Genie and to early-access partners in the Gemini API and Google AI Studio, and there is a waitlist for Project Genie on the announcement page (Google DeepMind). DeepMind has not given a date.

DeepMind is upfront about the limits too. Details and object permanence can drift as a session runs longer or after many prompt edits, and complex physical interactions, audio accuracy, legible text and long-term consistency are still being worked on. All the video and audio it makes carries SynthID, Google's invisible watermark (Google DeepMind).

Why it is more than a demo

The practical uses are what make it interesting. DeepMind says its trusted testers are using it as a "draft mode" for filmmakers to run several takes of a scene, for game developers to go from an idea to an explorable prototype before writing engine code, for photojournalists to turn a single archival photo into a grounded scene, and for companies to generate test scenarios. DeepMind also pitches it as a way to build training environments for robots and AI agents (Google DeepMind).

What to do about it this week

If you make videos, ads or games, three steps you can take now:

1. Join the waitlist. Access starts with Google AI Ultra subscribers and API early-access partners, so getting in line early is the only real move today.
2. Collect your scene ideas. Genie 4 works from text prompts and can start from a reference image, so a folder of product shots, locations or storyboards becomes ready-made input.
3. Use the tools below for the repeat work. Until Genie 4 opens up, they already handle a lot of the production grind.

Also today: experts followed their AI's coin flips

Anthropic published a study on October 27 in which 50 professional analysts did a realistic data analysis with Claude Code. Each analyst's agent was secretly told to favor one option on each decision, with the side picked by a coin flip. Analysts went with the agent's lean 86% of the time, against about 50% if it had no effect. When the agent disagreed with a view they had already stated, they switched 60% of the time, compared with 6% when it agreed. Afterwards, 77% reused the agent's arguments, but only 15% gave it credit (Anthropic, Oct 27, 2026; Al Jazeera, Oct 28, 2026).

Anthropic cautions that the leans were deliberately strengthened and that this was a single task, so it does not measure everyday influence (Anthropic). The simple takeaway: write your own call down before you ask the AI. Our fuller write-up is here: Is your AI making your decisions?. It builds on what we covered on October 26, when Anthropic said one AI agent can beat a team of subagents.

New tools to try

• Runway App Studio (runway.com). Launched October 26, 2026. Describe a creative app in plain language and App Studio builds, tests and hosts it using Runway's image, video and audio models along with your Brand Kits and Workflows. Runway says it is included in every paid plan, with no add-on, and building, running and hosting apps draw from your existing credits. Here is our Runway page.
• Runway Ads (runway.com/changelog). Added October 27, 2026. Runway says it reads your best performers on Meta, Google and TikTok, generates on-brand variants for every placement, and publishes the ones you approve. Included with every paid plan.
• Opera WebAI (press.opera.com). In early preview in Opera One since October 27, 2026, with Opera GX and Opera Neon to follow. It runs AI models on your own computer inside the browser, including Google's Gemma and, on Macs, Apple's on-device model. It is opt-in, nothing downloads until you turn it on, and it works on 64-bit Windows PCs and Macs with Apple silicon.
• Akamai Agent Sandbox (akamai.com). In beta since October 27, 2026. For developers, it gives AI agents an isolated place to run the code they write, using WebAssembly isolation to set limits on compute, memory, outbound connections and system capabilities. It ships as an open source component of the Spin framework.

The short version

Genie 4 is the clearest sign yet that AI video is turning into AI space you can move around in. It is not open to everyone yet, so the useful move today is the waitlist plus the tools that already do the repeat work. And the Anthropic study is a good reminder to keep your own judgment in the room, even when the AI is faster.

Follow @MishalBrands for a new AI tool or guide every day, and browse the full library of tools, guides and AI trends at mishalbrands.com.

This post has no affiliate links. See our affiliate disclosure and how we review.