Picture walking through a haunted 1970s Italian villa, clicking on objects to find clues, then typing a line of text to make it rain and watching the storm roll in. That is the kind of world Google’s Project Genie now builds with Genie 4, which Google opened to eligible Google AI Ultra subscribers on 18 November. Google AI Pro members can join a waitlist.
Genie 4 is a world model: instead of rendering a finished video, it generates the scene as you move through it in your browser. Google DeepMind first showed Genie 4 to testers in October.
What people are making
The new community gallery on the Project Genie page reads like a small arcade. “Villa Giallo” asks you to find all four clues in a room before the next one opens. “The Life of Ants” lets you live a day as a worker ant, sipping honeydew from aphids and carrying food home. “Where in the World?” drops you somewhere new and hands you a stamp for each correct guess.
Some of these came from early-access creators. Google says it built Villa Giallo with creative technologist Will aka BLVCKL!GHT and The Life of Ants with filmmaker Anand Gandhi. Other gallery worlds turn personal photos into spaces you can walk around (“Living Memory”) or make Gemini the live referee in a game of rock, paper, scissors played with objects on a desk.
How you build one
You can start from a text prompt or an image, or remix a gallery world by changing its structure, soundtrack or rules. A new Project Genie agent talks you through the idea and generates opening visuals plus background music to match the mood. Finished worlds can be published to the gallery for others to remix.
The headline trick is editing on the fly. While you explore, you type a prompt to shift the weather, drop in a character or redirect the action. For the first time the worlds also have sound, from ambient noise to characters who talk. That puts it a step beyond clip makers like Suno’s new lip-synced music video tool, where you watch rather than walk around.
Genie 3 stays in the model picker for longer runs: up to three minutes per generation (up from 60 seconds), worlds chained together, and scenes started from 360-degree panorama photos.
What it costs
Google’s AI plans page lists Project Genie under Google AI Ultra 20x, which costs $199.99 a month in the US, and says it is available in more than 140 countries. One footnote on the same page says any Ultra plan, so check your own plan before paying. Pro costs $19.99 a month but only gets you the waitlist, with no date given.
Where it breaks
Google DeepMind is frank about the limits on its Genie 4 model page. Genie 4 holds about a minute of continuous interaction. Details and objects drift the longer you play or the more you edit. It cannot reproduce real places with perfect accuracy, and legible text usually appears only if it was in your prompt.
We have not tried Genie 4 ourselves. The quality claims come from Google’s own demos, as Decrypt noted, not from independent testing.
Everything made in Project Genie carries Google’s SynthID watermark, a hidden mark that flags AI-made video and audio. That matters in Europe, where AI generators already on the market must mark their output by 2 December. DeepMind says Genie 4 builds on Gemini Omni; we covered Gemini Omni’s latest update earlier this month.
What to do this week
If you already pay for Ultra, open a gallery world first and remix it rather than starting blank. Change one thing at a time (weather, then a character), expect each Genie 4 session to last about a minute, and skip any scene that depends on readable signs. On Pro, join the waitlist on the Project Genie page.
Sources
Get the weekly email
The week's most useful AI news, a tutorial and an interview.
