• Hey Guest. Check out your NeoGAF Wrapped 2025 results here!

These AI Generated Interactive Environments Are Incredible

cmk74d7yh000xo40htja5iqrn
 
The title is the most impressive part of the video

Nothing in this video comes close to looking "almost real" lmfao.
 
Last edited:
I think thats pretty impressive, AI is such a great step up for making stuff easyer for us like the Internet or the Smartphone
 
Last edited:
Looks shitty and soulless.

The way GAF defends generative AI is surprising. Everybody here is negative and cynical about everything but for some reason AI slop is just fine.
 
Last edited:
Looks shitty and soulless.

The way GAF defends generative AI is surprising. Everybody here is negative and cynical about everything but for some reason AI slop is just fine.
Maybe it's just my perception, but I feel like the opposite is happening. To me, it looks like people here hate the idea of AI being implemented in games (and everywhere else).
 
This is generated by Google's stupidity named AI model, it'll likely be able to generate fully playable games in the coming years.
1. This is not a "videogame", it is a realtime video generator
2. Those models have consistency only for a few minutes unlike videogames
3. You cant create a controllable story or have consistent gameplay
4. This is so inefficent its beyond laughable, it wont scale

This will NEVER be able to generate fully playable games due to the way LLMs work!
You just fell for some marketing BS!
 
Procedural generation has existed for a long time and has been used in a lot of major games.
Procedural gen = fixed rules/math at load time, world never changes

AI here = neural net predicts every frame live as you pan camera, world gets invented in real-time from one pic

Not really the same is it?
 
Procedural gen = fixed rules/math at load time, world never changes

AI here = neural net predicts every frame live as you pan camera, world gets invented in real-time from one pic

Not really the same is it?
It's close enough in terms of output for the actual purpose of what you are doing which is building a game world to explore.
 
Once this technology advances, perhaps we will see a scenario in which the big pubs and devs lay off a lot of staff and you see team sizes shrink back down to the low hundreds for the big games from the high hundreds to thousands we see now.

But then at the same time AI will greatly reduce the barrier to entry and budgets and so you will see way more studios coming into the industry.

So these new studios and growing smaller studios will absorb those that have been culled in the initial AI guillotine in the larger teams.
 
It's close enough in terms of output for the actual purpose of what you are doing which is building a game world to explore.
It's not even fucking close

One's a reliable blueprint you can build a game on; the other's a dreaming brain inventing the world in real-time
 
Some of this shit looking like PS2 graphics

yes-this-is-an-actual-pc-game-screenshot-v0-wcikngx7py9a1.jpg
 
Genie 3 can generate dynamic worlds that you can navigate in real time at 24 frames per second, retaining consistency for a few minutes at a resolution of 720p.

Not a big AI fan but Impressive that its interactive moving through an environment a shame its limited to only a few minutes of consistency at 720p 24fps maybe things improve in the future but theres no game here. I can see it used in other use cases though.

The ability for developers to export the AI generated model data direct to unreal is an interesting tool.
 
Last edited:
Cool tech, but I'm not nearly as impressed as the dude who's talking on that video.

Honestly I hope this fad ends soon. I like videogames made by people that have put care and effort to balance the game, make interesting locations to discover and explore and created cool characters and settings. Leaving all that to the "imaginations" of a machine seems as soulless as the medium could ever get.
 
The title is the most impressive part of the video

Nothing in this video comes close to looking "almost real" lmfao.
It's not even fucking close

One's a reliable blueprint you can build a game on; the other's a dreaming brain inventing the world in real-time
For Google's 3rd major crack at it? This is very impressive. At worst, you can accuse it of looking like an asset flip game. Remember that first Will Smith eating spaghetti video? Look at the latest version of that.

On one hand, this is incredible.

On the other hand, we are well and truly fucked.
 
Some of this shit looking like PS2 graphics

yes-this-is-an-actual-pc-game-screenshot-v0-wcikngx7py9a1.jpg
compare PS2 graphics to PS5 graphics.

AI is in its early stage. It's only going to get better.

Current AI has a lot of flaws but if you showed me it 4 years ago it would blow minds. I mean I can spot an AI image or video from a mile away but it's come a long way.
 
For Google's 3rd major crack at it? This is very impressive. At worst, you can accuse it of looking like an asset flip game. Remember that first Will Smith eating spaghetti video? Look at the latest version of that.

On one hand, this is incredible.

On the other hand, we are well and truly fucked.
here is the Spaghetti "benchmark" video:

 
Interesting video. If current LLMs are at a 3DFX Voodoo 1 stage and the trajectory will go towards RTX 5090, then yeah LLMs are going to overtake everything.

But it is still unclear if it will scale effectively like that, considering HW and electricity requirements.

The video in OP algoed me towards this one, which is pretty fun. Generating 3D models so easily from 2D pictures, also generated...it is ridiculous. 100% of developers are going to be using this within the next two years.

 
Phil Spencer calls this game preservation, btw!

Satya Nadella driving to work rocking out to this
 
Last edited:
Classic internet nitpick. I called it neural net previously. so the 'dreaming brain' bit is clearly figurative language to describe how these models work
it's not a nitpick. It's fundamental to the sleight of hand that AI slopvocates use to try to make this stuff out to be more than it is. "it's dreaming", "it's thinking", "it has a brain" - no no no.

My point about procedural generation is that it is used to generate large game worlds. It's been around for a long time and works well. Does this technology work differently? Yes. Does it do things differently? yes. Does it make a better game world, as in, a better game world for exploration and doing what you do in a game? Well, I don't know. Maybe, maybe not. It hasn't generated any explorable game worlds yet. Nobody has made a game with this technology yet so we can analyze it properly.
 
Interesting video. If current LLMs are at a 3DFX Voodoo 1 stage and the trajectory will go towards RTX 5090, then yeah LLMs are going to overtake everything.

But it is still unclear if it will scale effectively like that, considering HW and electricity requirements.

The video in OP algoed me towards this one, which is pretty fun. Generating 3D models so easily from 2D pictures, also generated...it is ridiculous. 100% of developers are going to be using this within the next two years.


there seems to be some movement towards SLMs (small language models). probably more so recently with the rising hardware costs. SLMs will need to take up less storage space, maintain similar quality/performance to LLM, and run with lower RAM amounts. LLM's are in the hundreds (or thousands) of GB now in terms of size. you need a GPU with a similar amount of RAM to run them. I used to play about with some local models on my 5090 and even that struggles with 32GB VRAM on some of the higher end ones.

OpenAI's top local model is 120b which is a 64GB download. the entire model needs to fit in VRAM for best performane so you're looking at an RTX Pro 6000 just to run that. an RTX 5090 will struggle. You can run the 20b model just fine on a 5090 as it's 14GB in size. Qwen has a 235b model (142GB size) and Deepseek has models up to 671b (404GB size).

Local models don't really compare to the models that Google, OpenAI, Anthropic etc run in their web/cloud versions. ChatGPT 3 was about 350GB in size. latest models probably at 2-10TB.

SLMs are aimed at running on smartphones. so they need to be in the range of <50GB without eating too much into storage for the OS/user files etc.

Apple's dumb brick of an AI needs 7GB and that can't even tell you the weather accurately.
 
Last edited:
it's not a nitpick. It's fundamental to the sleight of hand that AI slopvocates use to try to make this stuff out to be more than it is. "it's dreaming", "it's thinking", "it has a brain" - no no no
It's a nitpick and you know it

Literally in dozens of papers they call it dreaming: learning behaviors purely by latent imagination dreaming in the compact latent space of the world model

It's not sleight of hand; it's the tech lingo
 
One would think and hope, but it's really hard to think about that until we see it. So, we'll see! Lol.
there is a bubble and it will burst.

right now companies are throwing ungodly amounts of money at AI but making very little returns and that's unsustainable. Eventually they'll be forced to create actual products people want/need and therefore create profit. the pop will reset and correct AI. we might find that there is an abundance of data centers and power plants. maybe an excess of hardware that needs to be shifted. prices will likely come down. AI isn't going anywhere but things will stabilize. maybe not this year or next but it will happen. this AI stuff is just like the GPU crypto boom.

but yeah we need to truly wait and see.
 
My point about procedural generation is that it is used to generate large game worlds.
Procedural generation = a specific seed always generates the same outcome for the whole virtual world
AI video generation = every single frame is like a new seed with the outcome changing randomly (and sometimes drastically)

Does it make a better game world, as in, a better game world for exploration and doing what you do in a game? Well, I don't know.
See this is where you make the mistake, this is not a game engine, a game world stays consistent throughout a game.
LLMs are not made to be consistent, they make stuff up as they go - like the writers of LOST - which then results in a mess.

It hasn't generated any explorable game worlds yet. Nobody has made a game with this technology yet so we can analyze it properly.
No one has made a game with this tech because its impossible.
The only thing this tech is useful for is neural rendering which might be added to game engines of the future!
 
Last edited:
Something that I have been mulling over for quite some time now. game developers ditching polygons for AI generated/ video and photo referenced assets. I thought about the concept of resurgence of FMV gaming, but this time, it would be aided by AI to create environments seen in the video featured above. What if someone were to make a Myst sequel or reboot using tech like this? Or a first person styled horror FMV game with real actors blended together with AI generated environments?
 
Seen some interesting videos on A.I making entire sets and characters with different existing franchises and series and some of them look quite cool and interesting. If it gets better than I'll definitely be way more interested seeing those compared to CGI stuff that looks more digital. At least some of those A.I could make things look fleshy and real while some look plastic.
 
Seen some interesting videos on A.I making entire sets and characters with different existing franchises and series and some of them look quite cool and interesting. If it gets better than I'll definitely be way more interested seeing those compared to CGI stuff that looks more digital. At least some of those A.I could make things look fleshy and real while some look plastic.

Oh it will definitely get better, and faster than we expect. Just a year ago people we're saying things like AI would never be able to fool the human eye because "evolved over thousands of years to spot inconsistencies." Fucking dead wrong.
 
Top Bottom