Google has opened Genie 4 to eligible subscribers of its Google AI Ultra plan, and members of the cheaper Pro plan can join a waitlist. Genie 4 builds a walkable 3D world from a typed description, lets you change it by typing while you are inside it, and adds ambient sound and speaking characters. Google also added a chat assistant that builds a world from a conversation, and a public gallery where people publish worlds and remix each other's.
A game stores its world and Genie does not
When Project Genie opened in January, game company shares fell because it looked like a way to make games without studios. The comparison breaks on how the two work. A game engine keeps a stable model of the world, so a door stays where the designer put it and every player sees the same thing. Genie calculates the scene frame by frame, reacting to your keys and your typing, while Gemini models plan in the background so the story keeps going.
The price of that freedom is memory. Objects and details can shift as a session runs longer, and spatial accuracy and readable text are still unsolved. A world that cannot hold still is a weak game. It is fine for a sketch, which you look at for a few minutes and then check.
The early users treat it as a draft
Filmmakers among the first testers step into a scene and run several takes, changing the camera, the light and the dialogue between them. Game developers go from an idea to something they can walk around in before writing any code. Photojournalists turn one archive photograph into a place. None of them is shipping the result. They are doing what a cardboard model does for an architect or a storyboard does for a director: making a rough version so cheap that they can make many.
What it changes for people who are not filmmakers
Anyone who presents a plan for a space is in the same position. Picture the head of a school department planning an open evening in the main hall. She photographs the hall in 360 degrees, starts a Genie 3 world from it, and types in a row of stalls along one wall and a crowd at the door. Then she asks two colleagues to walk in and say whether the entrance feels like a bottleneck. Comparing four layouts this way takes an afternoon, where drawings would take most of a week. A colleague who finds her world in the gallery can remix it with one sentence ("move the stalls to the left wall") and send it back.
The world will not tell her whether forty stalls fit. Measuring still does that, and it should come after the walk-through, because a layout that feels wrong is not worth measuring. In October we covered how machines rehearse in simulated worlds before the real thing; this is the same rehearsal in a person's hands, for ordinary plans.
My expectation is that within a year or two, the first draft of a shop layout, a classroom, a stand at a trade fair or a film scene will be a world that people walk through, not a slide with a picture. That makes the checking step more valuable, not less. The skill that gains is knowing what a sketch can tell you, which is how a place feels, and what it cannot, which is whether it fits.