How Google’s AI Created My Nintendo Game Rip-Offs
▼ Summary
– Project Genie is a new Google DeepMind AI research prototype that generates interactive 3D worlds from text or image prompts, now available to more users.
– The author used the tool to create worlds based on Nintendo franchises like Super Mario and Zelda, but Google later blocked some of these generations.
– The generated worlds are limited to 60-second, low-resolution experiences with significant input lag and a lack of objectives, making them poor as games.
– Google released this prototype to learn about new use cases, such as for filmmaking or education, but stresses it is not yet a polished, everyday product.
– The technology shows potential but is currently inconsistent and unreliable, being far worse than a handcrafted video game experience.
A new experimental AI tool from Google recently provided a hands-on opportunity to create brief, interactive 3D worlds, including some amusingly poor imitations of classic Nintendo games. Project Genie, a research prototype now rolling out to select users, represents Google DeepMind’s push into generative “world models” that can build virtual spaces from simple text or image prompts. While the technology hints at future applications in education, filmmaking, and robotics, the current experience feels more like a fascinating tech demo than a polished product.
The process is straightforward. You select a pre-designed environment or write a prompt describing the world and character you envision. After a short wait, Genie produces a thumbnail and then generates the full, explorable space. You get sixty seconds to navigate using typical keyboard controls, though the experience is rendered at a modest 720p resolution and roughly 24 frames per second.
The tool’s official demo worlds, like “Rollerball” and “Backyard Racetrack,” quickly reveal its limitations. Movement suffers from noticeable input lag, making precise control difficult. Worlds lack objectives, sound, or consistent logic, paint trails would vanish, or parts of a race track would inexplicably turn to grass. This inconsistency breeds a sense that you can’t trust the environment to behave predictably from one moment to the next.
The most entertaining use was testing its boundaries by prompting it to create worlds inspired by famous franchises. It successfully generated passable, if crude, approximations of settings from Super Mario 64, Metroid Prime, and The Legend of Zelda: Breath of the Wild. These AI-made knockoffs were good for a laugh but were essentially empty shells. There were no enemies to fight, puzzles to solve, or goals to achieve beyond basic movement, which was further hampered by the persistent control lag.
Not every attempt succeeded. A detailed prompt to create a Kingdom Hearts-style world was blocked from full generation, though the thumbnail preview showed characters strikingly similar to Sora, Donald, and Goofy. When asked about generating content based on copyrighted characters, a Google DeepMind product manager noted that Project Genie is an experimental prototype designed to follow user prompts and is trained on publicly available web data. The company monitors usage closely. Notably, access to generate Super Mario 64 worlds was later restricted, citing the “interests of third-party content providers.”
As it stands, Project Genie feels like a promising but early step. The combination of a strict time limit, technical hiccups, and a lack of engaging gameplay makes these worlds difficult to enjoy for more than a brief novelty. Researchers envision a future where such tools blur the lines between different media, but the technology has significant ground to cover. For now, these AI-generated worlds are nowhere near replacing the crafted experiences of traditional video games, and the genie of fully realized, AI-crafted interactive universes remains firmly in its bottle.
(Source: The Verge)
