Tech Briefing. Your daily dose of tech news — signal, not noise.
Wednesday, October 28, 2026
The blog companion to the podcast
DTBRIEF

Google’s Genie 4 Builds Worlds You Can Walk Through. Here’s What That Actually Means.

Google’s new world model generates photorealistic places you can explore, change with a text prompt, and hear — and it can hold the illusion together for two full minutes.

2 min
The longest consistent simulation Genie 4 can sustain — double Genie 3
2×
Genie 3’s one-minute limit, doubled in the new model
Real-time
Prompt edits applied live, while you explore the world
Listen to this story
Melody reads the top story · 3:06

1 What happened

Google DeepMind has introduced Genie 4, a world model that generates photorealistic, interactive environments — places you can walk through, not just watch. The model renders the world in real time, complete with synchronized sound, and you can change what you see with a text prompt while you explore: ask for rain, a different building, a new character, and the scene updates around you.

The headline number is two minutes. Genie 4 can keep a simulation consistent for that long, doubling the one-minute limit of its predecessor, Genie 3, while also improving physics and making characters look and move more realistically. In plain English: earlier world models made impressive clips that fell apart the moment you tried to interact with them. Genie 4 is built to stay coherent while you poke at it.

A “world model” is different from a video generator. A video generator predicts the next frame of a clip. A world model tries to understand the place itself — what’s solid, what moves, what happens if you walk behind a wall — and then renders whatever you look at, from whatever angle you choose.

2 Why it matters

First, this is the step from watching AI video to inhabiting it. A two-minute interactive world is still short, but it crosses a line: the output is no longer a recording, it’s an environment. The obvious first home is games, where a model that can dream up a playable level on demand changes what a small team — or a single player — can build.

Second, the same technology trains robots. A robot that has practiced in a thousand simulated kitchens, warehouses, and streets before it ever touches the real one is a safer, cheaper robot to deploy. Google has been explicit that world models are a core piece of its robotics work, and Genie 4’s longer, more physically consistent simulations make that training more useful.

A video you can walk through is a different medium from a video you watch. Genie 4 is the clearest sign yet that the industry is building toward the first one.

Third, it raises the stakes in the race to simulate reality. Film and virtual production teams can already block out scenes that behave like real places. The same capability, in the wrong hands, can build photorealistic places that never existed — which is why how Google limits access, and what it discloses about generated worlds, will matter as much as the model itself.

3 What to watch next

  • Public access Google has not said when — or how widely — the public will get to try Genie 4. Watch whether it arrives as a research preview, a developer tool, or a consumer product first.
  • Longer sessions Two minutes is the new bar. The next question is whether the following generation can hold a world together for ten minutes, an hour, or an entire game session.
  • The rivals Every major lab is chasing world models. Expect competing announcements, and pay attention to who ships a usable product rather than a demo reel.
  • The guardrails Photorealistic, interactive worlds are a powerful tool for misinformation. Watch what disclosure and watermarking Google builds in before wider release.
Original reporting

This post is based on Google DeepMind’s announcement: “Genie 4: a new generation of world simulation” — Google DeepMind, October 28, 2026. Figures are as reported by DeepMind.

Heard it in the briefing

This story led Episode 18 of the Daily Tech Briefing podcast — your daily dose of tech news in about 10 minutes, read by Melody, your AI anchor. Listen on Spotify or Listen on Apple Podcasts.