Google’s new AI model creates video game worlds in real time
https://www.theverge.com/news/718723/google-ai-genie-3-model-video-game-worlds-real-time
Google’s new AI model creates video game worlds in real time
https://www.theverge.com/news/718723/google-ai-genie-3-model-video-game-worlds-real-time
17 Comments
Pretty sure this is/was Midjourney’s whole endgame.
Summary: Google DeepMind has announced Genie 3, an AI “world model” that generates interactive 3D environments from text prompts, representing a significant upgrade from its predecessor Genie 2. The new model allows users to interact with AI-generated worlds for “a few minutes” instead of just 10-20 seconds, features better visual memory that remembers object placement for about a minute, runs at 720p/24fps, and includes “promptable world events” for dynamic changes like weather or adding characters. However, Genie 3 is only available as a limited research preview to select academics and creators while Google studies potential risks, and world models still face challenges like visual quality and object consistency that make them feel less polished than traditional video games.
How much water and power is being used to make these models? And is the effort to produce the model really worth 5 minutes of playtime?
I was only waiting for the first example to appear
This is going to be a massive part of future games. Imagine an adaptive game design system as Star Treks holodeck uses to allow characters to create and modify scenarios on the fly.
If it isn’t built in to consoles by the next generation, it will be in the generation following.
Everyone who is convinced this is going to be able to make videogames, I have a couple questions:
I am curious why you think people play videogames in the first place, and if you play them yourselves?
If you don’t then I can get how you would confuse stuff like this with “real” games; if you actually do play games, though, uh… Can you explain what the point of these would be? We’re always hand waving away the important stuff like *why anyone would play this* beyond the short lived novelty of it.
If your entire experience with games is a lens through which you can sell AI effectively, I suspect this question doesn’t even occur to you, but the folks who do, what’s your excuse?
I was on this exact board two years ago saying this would be possible. Now we need to get these to generate models and save its imaginations.
Absolutely crazy how far this stuff has come in just a few short years
Anti AI people who thinks AI is going away or won’t be used pervasively in every industry are in for a rough time
It’ll be feature length movies soon. In a couple years we’ll be going to the cinema and watching entirely generated movies. Except we won’t because that’ll suck.
At some point, Netflix is going to let you just generate whatever show you want instantly.
I hope it’s better than whatever it is Google Home uses
Watch lack of object consistency become a feature rather than a bug
This sort of thing will never enter production. Fully AI generated games via frame generation is stupid. It’s cool as a concept but that’s it.
What is much more likely is that ue5 or new, bespoke game engines get created that use AI to create artifacts on the fly like models or landscapes.
Whatever. As impressive as all this AI shit is, I’ve seen it, and it solves problems I don’t have.
No no but AI will never be able to make games. I asked it a question last week and it gave a wrong answer hurhurhur, surely it will never advance.
In my opinion. AI generating models inside a game engine makes far more sense than image generating a video game.
Welcome to the end state of humanity, the pleasure matrix approaches.
Very cool how my college major is now just completely automated in less than 5 years after graduating!!
#*Via Frame generation.
Yeah, this sort of thing will not catch on
It generates what it thinks the game is for a minute. ***It doesn’t generate any assets.***