Worlds You Can Describe: What the New Wave of World Models Means for Makers and Storytellers

Picture this. Someone types one sentence asking for a valley made entirely of stained glass, with a river like flowing blue glass and flowers like small glowing sculptures. A moment later the camera is gliding through it, and the valley keeps unfolding as it moves.

That isn't a scene from a film. It's one of the demo prompts Google DeepMind showed when it announced Genie 4 on October 27. I watched those demos the way I imagine many of you did: half amazed, half asking, "Okay, but what does this actually mean for people who make things?"

I want to try to answer that honestly. I'll talk about the people who build games, and in equal measure about the people who tell stories, teach, make art and bring communities together.

What a "world model" actually is

Most games you've played were built in an engine like Unity, Unreal or Godot. In those tools, everything is explicit. Someone writes the code, sets up the physics, models the assets and places every tree. The upside is precision: the world behaves the same way every time, and you control every detail.

A world model works differently. It learned from huge amounts of video, and it generates the world frame by frame, predicting what should come next based on what came before and what you just did. There's no hand-built level underneath. DeepMind described the consistency of Genie 3, the previous version, as an emergent capability, meaning it comes from what the model learned rather than from a stored 3D map.

Genie 4 takes that further. According to DeepMind's model page, you can change a world while it runs by typing things like "change the weather," "move the camera" or "add a new character." The model also generates its own sound: wind in the trees, crowds, even characters who talk back. TechCrunch reports that it can keep a scene consistent for up to about two minutes, double what its predecessor managed.

Google isn't the only one exploring this. World Labs' Marble turns text, images or video into 3D worlds you can edit and export. Runway's GWM-1 generates explorable worlds and conversational characters. Odyssey streams interactive video that you steer by typing. Decart's Oasis is aimed at training autonomous systems. Tencent's HY-World 2.0 is released as open source. NVIDIA Cosmos focuses on robots and other physical AI. They take different approaches, but all of them are working toward the same idea: describe a world, and step into it.

For the people who make games

If you've ever had a game idea at midnight, you know how far it is from "what if…" to something you can actually walk around in. Even a rough prototype can take days.

That's the gap world models speak to. The Verge notes that Google is pitching Genie 4 for prototyping game ideas. Imagine trying five moods for a level before breakfast: a foggy harbor, a desert market, a city of paper lanterns. You'd keep the one that gives you chills and build that one properly.

I want to be careful here. A sketch isn't a game. Games need rules, scores, saves and other players, and they need to behave the same way tomorrow as they did today. That's still the job of real code and real design.

For the people who tell stories

The creative side gets just as much from this, maybe more.

Filmmakers already use rough "previs" to plan shots before anyone builds a set or books a crew. The Verge reports Google highlighting pre-visualization for filmmakers, and Google's own post names filmmaking and journalism among the creative fields it could help.

Think about what a describable world could mean for a writer who wants to stand inside her novel's opening scene. Or a painter testing how a style would feel as a place you can walk through. Or a teacher. I'll frame this one carefully, because it's a possibility rather than something anyone has reported yet: imagine a class asking "what was it like on a 1980s train platform?" and exploring a scene together, then talking about what the model got right and what it got wrong. That second conversation might be the most valuable lesson of all.

The honest part

This is where I think we owe each other some clear thinking.

It's a preview. Genie 4 is with trusted testers for now. Google says it's "coming soon" to some subscribers and developers. Most of us can't use it today.

Worlds drift. DeepMind's own limitations section says details and object permanence can shift over time, especially in longer sessions or after many edits. Sessions last minutes, not hours.

Shared worlds are hard. DeepMind says modeling many independent agents in one environment is still an open research challenge. That matters a lot for anyone who dreams of multiplayer.

Control is fuzzy. UploadVR made a sharp point about Genie 3: getting a model to match the exact details you ask for is hard, which is one reason traditional rendering isn't going away.

Big questions are still open. We don't yet know what these systems will cost to run, and DeepMind hasn't published things like resolution or benchmarks for Genie 4. Questions about training data and creative ownership deserve real answers too.

None of that takes away from how exciting this is. It just reminds me that the tools we can share, keep and control still matter enormously.

What you can do today

This is the part I care most about, because imagination doesn't need to wait for a waitlist.

At VTC, we build for people who want to turn ideas into places where others can gather. On vtc.com, you can create your own Spaces and Networks, free to start: a home for a story world, a classroom, an art collective, or a community of players. Our AI app builder lets you describe an app and shape it into something real. And Vee, our AI guide, is there for you along the way. She helps you find your footing, answers your questions and makes it easier to get from idea to something you can share.

If games are your thing, LobbyNet on miskio.com is VTC's games portal. You can play lots of games there and create your own with VTC's SDK and developer guide. These are real, persistent experiences that your friends can come back to tomorrow.

The way I see it, world models are a new kind of sketchbook. The places where we build together, play together and keep coming back are where those sketches become something people can live in.

Start with a sentence

Every world, whether it's generated, coded, painted or written, starts the same way: someone describes it.

So here's my invitation. Write down the world you'd describe if you could. Then bring it to vtc.com and give it a home: a Space for your story, a Network for your community, or a game on LobbyNet. I'd love to see what you imagine.

— Mark