Project Genie lets you generate new worlds 60 seconds at a time, but only if you pay for AI Ultra.
See full article...
See full article...
Google Project Genie lets you create interactive worlds from a photo or prompt
- Thread starter JournalBot
- Start date
* sniffs *
Smells like desperation to me.
So, has AI made a profit yet? No? More burning of VC revenue and the goodwill of any corporate tie-ins created to keep the AI hype train going, speed-running its demise.
On a more serious note, does this offering have a subscription rate that covers the actual costs of its use? If not, yeah, it's adding to the deficit of AI use. Google being supported by MORE than just AI can stand to have write-downs on some of its operations to reduce a tax burden from egregious income levels from advertising and selling data, so that kind of tracks as a loss leader.
That may be why they limited it to people with current subs only. Open this to the general public for free, and they'd be bleeding red all over hell.
Smells like desperation to me.
So, has AI made a profit yet? No? More burning of VC revenue and the goodwill of any corporate tie-ins created to keep the AI hype train going, speed-running its demise.
On a more serious note, does this offering have a subscription rate that covers the actual costs of its use? If not, yeah, it's adding to the deficit of AI use. Google being supported by MORE than just AI can stand to have write-downs on some of its operations to reduce a tax burden from egregious income levels from advertising and selling data, so that kind of tracks as a loss leader.
That may be why they limited it to people with current subs only. Open this to the general public for free, and they'd be bleeding red all over hell.
Upvote
35
(49
/
-14)
At $250/month, I expect to be able to move into the damned thing. At least part time.
Upvote
50
(51
/
-1)
Is there a reason they use AI to generate video instead of AI to generate assets which are fed into a more traditional rendering engine?
It would seem to me that solving for smaller parts of the whole would be cleaner and quicker than simulating the complete end result. Build the holodeck rather than a short YouTube video about the holodeck, in other words.
It would seem to me that solving for smaller parts of the whole would be cleaner and quicker than simulating the complete end result. Build the holodeck rather than a short YouTube video about the holodeck, in other words.
Upvote
71
(71
/
0)
Reminds me a lot of AR, like how in Pokemon Go you can spawn a pokemon "into the real world". Its a cool thing that people play around with once and then never again.
I think this looks equally as neat, but it lacks depth, so no one will use it more than once or twice.
I think this looks equally as neat, but it lacks depth, so no one will use it more than once or twice.
Upvote
15
(15
/
0)
ubercurmudgeon
Ars Tribunus Militum
I can create a virtual shiab inu, and stare at its anus for 60 whole seconds, as it walks around a slightly eerie rendition of an alpine meadow? What a time to be alive! And it only costs $250 a month, enough energy to run a small town, and potentially the viability of the profession of game designer? Bargain!
Upvote
54
(63
/
-9)
I can't see this ever becoming capable enough to generate a video game. It would only be interesting as something new that we haven't seen before. It might become pretty fun, with really strong emphasis on "might."
That resource consumption, though. The level of hardware they must be throwing at this to get it to work at all. It boggles the mind.
That resource consumption, though. The level of hardware they must be throwing at this to get it to work at all. It boggles the mind.
Upvote
10
(11
/
-1)
Upvote
13
(15
/
-2)
This made me remindo of the rotating Beethoven head demo of IRIX.
Neat indeed. Worth a 50K USD workstation? Nah. (surely not at the end of the IRIX cycle, for sure - you could get already a quadro). What is the point? To prove that "you can"?
And whenever I power my Octane on, the first thing I do is take Beethoven for a spin. At least I am not paying 250 USD/month for it. =)
Neat indeed. Worth a 50K USD workstation? Nah. (surely not at the end of the IRIX cycle, for sure - you could get already a quadro). What is the point? To prove that "you can"?
And whenever I power my Octane on, the first thing I do is take Beethoven for a spin. At least I am not paying 250 USD/month for it. =)
Upvote
3
(4
/
-1)
Actually, I want to see what this generates from a screenshot from Wolfenstein 3D or Doom.
Upvote
8
(8
/
0)
and 3D artist. my brother graduated university in 2022 and hasn't been able to find a single opening since. in Montréal, which has a bunch of studios. and it's not like he's choosy - he'd even go work for Ubisoft ! half his cohort already pivoted.
Upvote
14
(14
/
0)
I've wondered the same in a couple of occasions. It seems they think consistent works models are important, probably for robotics. The visuals are a way to test it, maybe with some selling opportunities.
I agree with you, being able to chunk a very complex problem would seem like a critical skill too.
Upvote
9
(9
/
0)
Reminds me of the scene in Inception when the old man is explaining to Cobb that "for them, the dream is the reality". Seems like something lots of folks here could relate to.
Upvote
0
(1
/
-1)
Maybe it's just me, but something doesn't smell right about this article. Maybe it's because something also didn't smell right about this Ars article.
Upvote
-17
(0
/
-17)
I'd agree.
The big potential commercial use case for this is robotics/autonomy. Right now, a large amount of effort goes into creating simulation environments. I can see the potential appeal of this technology, and the desire to create entire, coherent worlds. If AI could automatically generate simulation environments, it would save huge amounts of effort.
However, until AI can create longer, repeatable, fully coherent simulations with different sensor types (point cloud, multiple cameras, IR) it's no more than a cool demo. The physics models for the autonomous systems need to be able to interact with the world, its objects and surfaces. The sensor data needs to be returned in a form that closely mimics the real sensors. Engineers need to be able to run the same simulation dozens of times, tweaking the system and observing the results.
This can all be done right now with 3D environments and detailed sensor models. Just generating assets (or entire scenes) to go into those simulations would be a quick win. If the AI could generate detailed models on-demand that respond in a realistic way to different sensor types (e.g. having textures/shaders for visible light, IR and Lidar intensity), then that would itself be a valuable product. However, if they were just rough models that need to be tweaked to do anything but look pretty, then there would be no real advantage over the current approach of buying off-the-shelf assets and putting in the effort to get them up to scratch.
Last edited:
Upvote
4
(5
/
-1)
AI lets you "create worlds" like for example the White House uses it to create a propaganda world making the people they arrest look worse, and making Trump look better.
AI is the "socialist realism" of the modern world. It has become a propaganda tool first, and a novelty second, and a legitimate business plan not at all. The people working on it should be ashamed of themselves.
AI is the "socialist realism" of the modern world. It has become a propaganda tool first, and a novelty second, and a legitimate business plan not at all. The people working on it should be ashamed of themselves.
Upvote
2
(8
/
-6)
It does seem obvious to me too, get these things generating assets and worlds in Unreal Engine, you get coherency, physics, rendering etc virtually for free.
I wonder though if the issue is training - where would the data come from for the AI to learn to do that? Video is easy, they've got zillions of hours of YouTube etc. Teaching it how Unreal works (not how the output looks, but what to do to get that output) might be a lot harder. Feed it the manuals and some sample projects? I dunno.
Upvote
5
(5
/
0)
Reminds me of "The City on the Edge of Forever" from star trek.. Imagine a game that you pass through the portal to experience the world you chose.
Upvote
2
(2
/
0)
You start with a single triangle. Add more triangles at random, if it looks more like the object you give the AI a virtual cookie, if it doesn't you have it throw out the changes and add more triangles at random until it does. Everything that exists can be represented by a number of triangles greater than 1 and less than some upper bound I haven't quite pinned down yet. Once we do finally figure out that upper bound the software will run a looooot smoother, what with not having to keep all those extra triangles in memory all the time until we need them.
Upvote
4
(4
/
0)
Upvote
1
(1
/
0)
almostinsane
Ars Praetorian
So much $ and power wasted for a nothing app/product/turd. Seriously, WTF is this? Not exactly something that helps get the negative AI opinion reduced in any capacity.
When El Goog announces another round of layoffs, the majority better come from groups and divisions that waste their time developing this garbage.
When El Goog announces another round of layoffs, the majority better come from groups and divisions that waste their time developing this garbage.
Upvote
-4
(2
/
-6)
Not having ever used any of these “tools”, I’m trying to come up with an actual use case (beyond just playing with it) - at least one that justifies the cost. While I don’t expect those here to offer a valid one - there’s obviously a very healthy skepticism of the entirety of “AI” here - does Google offer one?
Upvote
-5
(1
/
-6)
I need way more flexibility and control for that amount of money to make it useful.
Upvote
3
(3
/
0)
It’s clever, and kinda cool, but by the end of the promo video I was already bored. What planet-destroying fancy is next…
Upvote
0
(2
/
-2)
So it's exactly the thing that it is not, technically, though.
It's just a video generator? Nothing to do with the world? Are we going to swallow the term "world model" because these incompetents require it for marketing?
Upvote
0
(3
/
-3)
Wow. This is pretty useless as is, but as a technical demonstration, it's very impressive to me.
But that's the thing with "AI" right?
Everything is impressive, but not really useful.
But that's the thing with "AI" right?
Everything is impressive, but not really useful.
Upvote
2
(4
/
-2)
Upvote
4
(4
/
0)
This strikes me more a foundational research. It's not intended to be a final product. It is intended to build out the internal knowledge and breakthroughs for better options.
The publication of this research probably has to do with stock prices and demonstrating their skills in the market.
The publication of this research probably has to do with stock prices and demonstrating their skills in the market.
Upvote
7
(7
/
0)
I wish tech companies would publish their internal project approval documents when they release a new product. Is there one? How did someone justify a team working on this? There has to be some sort of project qualification mechanism...right? Its neat but functionally pointless.
Upvote
-1
(2
/
-3)
Not today, not tomorrow, but someday we will have game (?) or 'experience' -engines that use something like this as a front-end. I don't see it happening until affordable consumer GPU/CPUs can provide the (bulk of) compute though. Not any time soon then...
On a base level, this seems almost like video generation coupled with "video2video" stable diffusion/comfyui workflow. You can do something like this at home with sufficiently powerful GPUs, albeit much lower resolution, low FPS (if you want real-time), occasionally incoherent, and non-interactive. This "project genie" seems far beyond that. It has the added complexities of interaction, decent resolution, decent frame rate, and - above all - somehow maintaining a cohesive image that doesn't lose context of what it has already displayed (evidently storing a cache or map of what has already been generated). My point being that there are a truckload of systems and logic to make all the necessary plates keep spinning for this illusion. It is incredibly impressive but equally impractical.
On a base level, this seems almost like video generation coupled with "video2video" stable diffusion/comfyui workflow. You can do something like this at home with sufficiently powerful GPUs, albeit much lower resolution, low FPS (if you want real-time), occasionally incoherent, and non-interactive. This "project genie" seems far beyond that. It has the added complexities of interaction, decent resolution, decent frame rate, and - above all - somehow maintaining a cohesive image that doesn't lose context of what it has already displayed (evidently storing a cache or map of what has already been generated). My point being that there are a truckload of systems and logic to make all the necessary plates keep spinning for this illusion. It is incredibly impressive but equally impractical.
Functionally pointless today, but think about it from a dystopian, greedy, business perspective: There is big marketing potential where a generated experience can have visuals (be it ads, product placement, psychological cues, etc) tailored to individuals as they watch or interact in real time.
Upvote
0
(1
/
-1)
I know this is unpopular opinion here but I just don't get it.
This is genuinely cool - I'm having a lot of fun playing around with it. If you don't like it, don't use it.
This is genuinely cool - I'm having a lot of fun playing around with it. If you don't like it, don't use it.
Upvote
0
(3
/
-3)
Google has always prided itself on letting its developers work on things that are merely neat while deferring judgment on whether they turn out to be pointless. The hope is that they'll be the only company that chances upon an amazing application of something that everybody else had dismissed as pointless — albeit with the irony that Google published word2vec way back in 2013, which ws fairly foundational to modern LLMs even though it is now outdated, but seemingly without anybody there realising the next step that would unlock a money bubble.
Upvote
3
(3
/
0)