There a wait to use this but very interesting. Working demo. Magica 2
Deepmind has been working on this for a bit also but havn’t seen anything released yet. called Genie 3.
The fact that they can’t support even accessing a pre-defined instance and there’s a 12+hour wait for it to respond to the sample prompt points to this being another example of an AI generating an “interactive scene” in the same way a generative AI creates an image. This is never going to be a realistic way to create games with anything like the models we have right now.
Nvidia is also working on this also. Difix3D+: Improving 3D Reconstructions with Single-Step Diffusion Models
Yes, I know there are several examples of exactly this sort of thing. That was implied in what I wrote. The point is that they aren’t ever going to be game engines. They make a snippet of game-like interaction in a 3D space. There is no linear improvement towards them becoming games.
1 Like
Still early stage but I can see it evolving to something more contextually driven. What’s really awesome this is pure frame generation no concept of polygons or 3D models. The output of the network model are the pixels on the screen.
You would probably have whole genres of feature-length films more or less entirely cut from AI-generated footage first. That use had the advantage of non-interactivity and manual post-processing.
Though they do share much in common.
i’m actually far more interested in ai game engines for npc dialogue and strategy
3 Likes
The future for digital entertainment is looking very bleak…
2 Likes
Why do you think that?
If you’re about to reply being concerned about “generic AI slop”… its very early days yet.
I will chime in with an answer to that, though not the one that said it.
The main issue is creativity. Not in the traditional sense, you will get something we would call creative, but it will all average out to the same answers, because the input we give it is all the same.
A story that details this is a professor gives his class an assignment to come up with an interesting project topic using AI and then turn it in the next day. The students do. Then the professor polled them on how happy they were with their topic, and everyone was mostly happy with it.
Then the professor showed them the topics that everyone had turned in, things like, “Effects of Microplastics in the Water supply,” “Effects of Microplastics in the Food supply,” “Efforts to combat Microplastics in nature.” etc.
The students became unhappy with their topics after this.
(These are examples, but not the original ones, but the idea still gets across.)
AI is only giving you in a particular range, and the same and similar answer are what you will get given the same input.
There are other parables that express the same thing, like the writer that needed an ending, and was on a deadline, so had an AI write an ending for him, got his script optioned, and a month before the movie was shown, another movie was shown with the exact same twist in their ending.
This property is great for coding, the output remains rather consistent, but for creative work, you want to explore the edges, not stay in the middle path, which is what AI would do.
1 Like
i think it could work, but not just using generic AI for the entire game.
each NPC would need goals, a personality, etc.
this would necessitate basically a unique prompt for each NPC.
I think the trap a lot of people fall into is assuming LLMs or whatever other current ML/AI we have can replace their creativity.
No.
Its just another tool in the toolbox imho, and its not going to replace human creativity for this sort of thing any more than the camera eliminated artistic endeavour or killed painting.