On January 29, 2026, Google opened Project Genie to Google AI Ultra subscribers in the United States, aged 18 and up. It is an experimental research prototype from Google Labs, and the first public-facing product built on Genie 3, DeepMind’s general-purpose world model. Give it a text prompt plus generated or uploaded images, and it produces an interactive 3D world you can actually move through, generated as you go.
The shockwave landed the next day. On Friday, January 30, videogame stocks sold off sharply in afternoon trading: Reuters reported that Unity plunged nearly 24% intraday to around $29.75, its worst one-day drop since 2022, while Roblox and Take-Two Interactive, the publisher behind Grand Theft Auto, slid alongside it — a decline The Verge tracked company by company. How does a research prototype capped at 60 seconds per generation knock an entire industry off its valuation? The answer says as much about world models as about the game business.
Genie 3: From Research Preview to First Public Product
A world model simulates environment dynamics: it predicts how a scene evolves and how actions change it. DeepMind previewed Genie 3 in August 2025 as a general-purpose world model and tied it explicitly to its AGI ambitions — general intelligence has to cope with the diversity of the real world, not a single closed environment like chess or Go. Project Genie is the first consumer surface for that research: a web app that pairs Genie 3 with Nano Banana Pro for image generation and Gemini. It launches in the US first, pointedly excludes Google AI Ultra for Business accounts, and more territories are planned.
Three Modes: Sketch, Explore, Remix
The product is organized around three capabilities. World sketching: you build a character and an environment from text plus images, with Nano Banana Pro handling preview and fine-tuning of the world image, plus a choice of first-person or third-person perspective and movement styles like walking, flying, or driving. World exploration: the model “generates the path ahead in real time” — this is not a pre-rendered static 3D scene but a world that grows segment by segment as you act, with an adjustable camera. World remixing: you modify existing worlds through their prompts, browse a curated gallery or spin the randomizer for inspiration, and download video recordings of your sessions.
Gaming Stocks Take the Hit
The market reaction on January 30 is more instructive than the product itself. Reuters reported videogame shares falling hard in afternoon trading after Alphabet’s announcement; The Verge noted Take-Two, Roblox, and Unity all closing down on Friday. The logic is not hard to reconstruct: the moats of engine makers (Unity) and user-generated content platforms (Roblox) sit exactly where the “one prompt, one playable world” narrative lands. Whether the market was pricing disruption or panic is still being argued — on Hacker News, one of the hottest threads debated whether GTA VI’s expected revenue could offset Take-Two’s pressure at all.
The 60-Second Cap and Other Limits
Google is candid about what does not work yet: worlds may not look photorealistic, may drift from your prompt or uploaded images, and may violate real-world physics; character control can be imprecise and laggy; each generation is capped at 60 seconds; and “promptable events,” a capability shown in the August 2025 Genie 3 announcement, are not included. In other words, today’s Project Genie is a research prototype, not a game-engine replacement. What actually matters is that it turns “interactive world generation” from a paper into a web page any AI Ultra subscriber can open — with official language about expanding to more territories and broader access over time.
How Developers Should Read World Models
Three practical judgments. First, the boundary between world models and video generation is dissolving: Genie 3’s output is essentially actionable video, which makes it a new source of material for games, animation, and robotics simulation. Second, the cost structure of content platforms is being rewritten: when the marginal cost of an interactive scene approaches the cost of one API call, UGC platforms have to re-prove the value of their matchmaking and community. Third, do not wire a 60-second prototype into a production pipeline — treat it as a validation tool for rapidly testing level designs, scene prototypes, and interaction hypotheses. For a wider lens on the industry’s signals heading into 2026, see our opening-of-year outlook; along the world-model track, Project Genie is the first checkpoint opened to the public.
Sources
- Project Genie: Experimenting with infinite, interactive worlds — Google
- Videogame stocks slide on Google’s AI model that turns prompts into playable worlds — Reuters
- Video game company stock prices dip after Google introduces an AI world-generation tool — The Verge
AI-assisted summary compiled from the sources above, reviewed by a human before publishing.