We're headed to text-to-fullyAnimated3DScene while people are still debating ethics of LLMs text responses. We're not ready for the future that is fast approaching.
I'm very sceptical if this is going to actually manifest. My wife and I have been digging into the text to 3D topic for a while now and the problem seems to be that there is just not enough correctly labeled data and you have to deal with the complexity of having to generate a mesh in the third dimension. There is just a vast amount of 2D pictures with descriptive alt texts or other textual context on the Internet which is simply not true for 3D models.
A lot of papers seem to try and extract 3D data from 2D pictures since the dataset is just so much better, but that hits obvious limitations. And we aren't even talking about any animations yet, which anyone who ever tried making them by hand knows, add a huge jump of complexity because you are not just morphing individual vertices, you actually want them to move consistently or else the faces are going to start doing weird glitchy things. Generating 3D stuff is less like generating pixel-based images and more like generating an SVG where the shapes actually make semantic sense, but a lot more complex. (Which, as far as we have found, no AI has managed so far either.)
Sure, you'll start seeing animated faces, architectural plans, and you'll also see bone system, based animations for humanoids (humanoid 2D video to 3D bone animation AIs exist today), but don't expect to be able to prompt your way through creating a Pixar movie.
You don't have to generate a mesh token by token like you generate text. You can map the latent space to parametrized 3D models, or generate SDF, or a NeRF, or something else.
Unfortunately, humans tend to react sufficiently strongly only when something truly tragic happens all of a sudden. Things that creep in don't usually elicit such a response.
Soon enough the only differentiator for videogames will be the marketing budget and brand power, soon some indie games will have AAA graphics and animations but zero sales as the public gets flooded with offers and become numb to new offerings, instead the differentiator would be what games your friends play (e.g. to play online together) and similar social dynamics, and not much the technical aspects; a world where everyone can paint the Mona Lisa makes it less valuable, but DaVinci would still be a solid brand for historical reason, but new painters not so much.
I think the differentiator will be a novel experience and that can only created by novel human characters. The last kind of the hill, standing over the kill, will be a author. What a ironic circle..
There is no evidence to support such claim, might as well the last king of the hill will be those who can find one amazing story among the miriads created by AI and pretend it was created by him/her.
The real division between games and games will, as it always has been, be if the game is fun or not. Nothing else matters. If it's not fun, it'll flop and disappear, some quicker than others. But if it's fun, people will continue to play it.
This doesn't ring true, just for example "Among Us" sales only skyrocket after a famous streamer picked it up, the sales were below average before that specific event that happened by mere luck, and it was just as fun before he made it popular.
Graphical fidelity isn't the problem today already. You have Epic Megascans, online asset libraries where you can buy crazy high quality models for dirt cheap and whatnot. However, the new visually high quality games that come out (and have a download size of 20+ GB for a top down game) don't necessarily do well despite a marketing budget. In contrast, games with very little budget but a unique (and sometimes crappy) artstyle and compelling gameplay loop can do crazy well if they scratch an itch and/or are good at talking to the press. I mean, look at Brotatos artstyle... just... look at it.
Your assertion fails even the slightest scrutinity, I* was able to create this extremely beautiful piece (https://i.imgur.com/4AOHe6A.png) just minutes after installing some stable-diffusion-GUI and I'm neither experienced or driven, AI is not just a tool, its way more close to the definition of robot (or bot) than the definition of tool; and its just beguinning, those models are gonna get better and better, including a better understanding of the nuances of what its being asked for it based on context clues (e.g. previous prompts, tags, etc) making it even less important to become "experienced" or "driven" as the models themselves favor those audiences.
I...suspect your Mk1 discriminator could use a few thousand more epochs of training. Those SD kids are spending hours trying to pull a best seed for prompts, and then not getting much interests(public and financial) back.
Apparently these image generators are starting to be used in those industries for ideas and mockups - by first drafting conceptual arts using generators, and then tasking humans to completely redraw everything by hand. Final images contain nothing generated, so no copyright issues, no uncanny valleys like found in that image, no professional job security concerns. That's what it's doing. Supercharging the best, giving false hope to the rest.
