Yes, it’s another article on AI this week. Google released Project Genie 3 to its Google AI Ultra subscribers in the US at the end of January, triggering a fall in the share prices of several big game studios and associated companies, such as Unity, Nintendo, Roblox, and Take-Two. But what is Project Genie, how does it work, and will it really disrupt the game industry?
What is Project Genie?
In Hazelight’s co-operative game ‘Split Fiction’, players take on the roles of Mio and Zoe, two young women who sign up as subjects in an experiment by a company called Rader. The company has created The Machine, which (spoilers!) is able to harvest and extract Mio and Zoe’s creative ideas and turn them into simulated worlds that players traverse, solving puzzles along the way. The game is a not-too-subtle-metaphor for AI content creation and the risk of plagiarism and copyright infringement by large language models. (Project Genie users immediately started to use the tool to generate worlds that look very close to Nintendo’s Pokémon and Zelda franchises, for example.)
Until recently, there wasn’t a real-world equivalent of the Machine that you could try. Enter Project Genie: an AI ‘world model’. Unlike large language models, which are trained on large datasets and used to predict the next token in their natural language outputs, world models are trained on mathematics and data representing the physical world. The model then generates a simulation of the physical world which responds in a realistic and consistent way to inputs. Under the hood, Project Genie is powered by Google’s AI models Nano Banana Pro and Gemini.
What this means in practice is that you can give Project Genie a natural language description of the world you want it to generate (or upload an image to use as a base) and the character you want to control within that world, and it will generate a photorealistic world that you can then explore.
Will Project Genie disrupt the game industry?
From the footage I’ve seen of Project Genie’s output, it is able to generate visually interesting settings, with the world generally reacting realistically and smoothly to the character’s position and motion. But like many image-generation models, it was not able to handle certain difficult prompts; Matt Binder’s article at Mashable.com incorporates many examples from users on X, including examples where the simulated world broke consistency when simulating a bathroom with multiple mirrors on the walls; there were extensive visual artefacts in another world based on the Walk of Fame on Hollywood Boulevard in Los Angeles; and text appeared mangled in an environment meant to look like an old Blockbuster video rental store.
Even if the world generation were pixel-perfect, this isn’t sufficient to create a compelling game. Games vary greatly in the experiences they deliver to players, but one common framework to break down what makes a game is Jesse Schell’s elemental tetrad: mechanics, aesthetics, story, and technology. Project Genie’s game worlds are somewhat impressive in terms of a mechanistic game world that looks great, but they don’t deliver on the story.
Now, potentially Project Genie could be coupled with a large language model to generate an overarching narrative, characters, quests and so on, so that you could explore a fully realised game environment generated in real time. However, players appear to strongly dislike AI-generated content in games, which often falls far short of the standard of quality found in modern video games, so games created with AI assistance would need a lot of human intervention to make them appealing to players. And in terms of technology, the Project Genie generations are currently limited to generating 60-seconds-long experiences, so this would also need to be extended. So the risk of disruption to game studios may be overblown.
Robot Wars, anyone?
AI systems are not yet able to handle robots moving autonomously in the real world; the recently released International AI Safety Report 2026 observed that AI systems “cannot yet integrate with robotic components to perform basic physical tasks such as housework… Progress on digital tasks has also proved difficult to translate into robotics, where the complexity of the physical world introduces new challenges”. World models like Project Genie therefore represent, as Google describes it, “a key stepping stone on the path to AGI” – a necessary advance in the development of AI robotics – as tools to help train AI agents to navigate 3D environments.
This could potentially have military applications, as Amir Husain points out in an article for Forbes.com, where he predicts that such models could accelerate the deployment of autonomous weapons systems, allowing them to be trained on simulated data before deployment on the battlefield, as well as enabling more general capabilities like simulating how a military attack might play out in the real world. This wouldn’t be too surprising: games have a long heritage of being used as representations of aspects of warfare, chess being a commonly cited example, another one being the use of flight simulators to train drone operators. The genie, as they say, is out of the bottle.
Further reading
On Project Genie
- Google, January 2026. Project Genie: Experimenting with infinite, interactive worlds. https://blog.google/innovation-and-ai/models-and-research/google-deepmind/project-genie. Google’s blog on Project Genie.
- Jon Martindale, January 2026. Google’s Project Genie Tool Lets You Build Virtual Worlds for Training or Just Fun. https://uk.pcmag.com/ai/162878/googles-new-ai-tool-lets-you-build-virtual-worlds-for-training-or-just-fun. Reports the release of Project Genie and offers some comments on whether gamers would engage with AI-created games.
- Amir Husain, January 2026. Why World Models Like Google’s Project Genie Work. https://www.forbes.com/sites/amirhusain/2026/01/29/why-world-models-like-googles-project-genie-work/. Offers an explanation of how world models work in terms of a neural network compressing physical rules into general principles.
- Matt Binder, February 2026. 10 cool examples of Project Genie, the AI world model that sent video game stocks diving. https://mashable.com/article/10-cool-examples-google-project-genie-ai-tool-sent-video-game-stocks-diving. Provides examples of Project Genie-created worlds generated by users.
On game development
- Bryn Gelbart, March 2025. Split Fiction Is Peak Co-Op Gaming, But Skip the Cutscenes. https://www.esquire.com/uk/culture/a64067309/split-fiction-game-review/. Review of Hazelight’s game Split Fiction, which describes the game as “a story about AI, large language models, and digital plagiarism in art”.
- Nick Yee, December 2025. Gamers Are Overwhelmingly Negative About Gen AI in Video Games, but Attitudes Vary by Gender, Age, and Gaming Motivations. https://quanticfoundry.com/2025/12/18/gen-ai/. Reports a December 2025 survey showing that 85% of gamers held negative attitudes to the use of generative AI in video games, especially where it would be used for creative elements of the game.
- Bhagpuss (nom de plume), January 2026. Yesterday Project: Gorgon, Today Project Genie. https://bhagpuss.blogspot.com/2026/01/yesterday-project-gorgon-today-project.html. A blog post arguing that game industry disruption from Project Genie is overstated, because even if AI-generated games were technically amazing, there wouldn’t be much of a market for them – players would prefer games that many other players are playing.
On robotics
- Yoshua Bengio et al., February 2026. International AI Safety Report 2026. https://internationalaisafetyreport.org/publication/international-ai-safety-report-2026. Annual review of major risks and mitigations of AI systems.
- Amir Husain, January 2026. World Models Like Google’s Project Genie Can Enable Autonomous Weapons. https://www.forbes.com/sites/amirhusain/2026/01/31/world-models-like-googles-project-genie-may-enable-future-hyperwar/.
- Johanna Urbancik, December 2025. Drone combat at home: Simulator lets players step into the frontline. https://www.euronews.com/2025/12/10/drone-combat-at-home-simulator-lets-players-step-into-the-frontline. Describes a game called ‘Ukrainian Fight Drone Simulator’, based on the simulator used to train Ukrainian drone pilots.