So....that big AI image or video to 3d mesh never materialized, did it? Its mid april. We should all be free of modelling by now. All this big AI talk, but nothing delivered. No image to model, no video to model, no ai texure, no ai uv, and github copilot stinks on ice.
>>980511Scary things rolling out in computer vision atm and 'text2music' is now a thing doing to music what generative AI did to images last year.'Agent AI's' are successfully accomplishing a lot of white collar work now. I dunno why you're in such a rush to get irrelevant anon, it'll happen in a heartbeat. Many within the field believe AGI can exist before we're even in the 30's given recent developments. They're currently building a lot of compute infrastructure at the moment and the roll out of GPT 5 later this year risk being the point when civilization at large and not just enthusiasts who follow the field realize just how F'ed we are.These things are not standing still at all anon. It's more about how the fantastical has become so common place you don't hear about it on the news as the breakthroughs are coming at such a rapid pace, way it looks from where I'm standing we're just sleepwalking into human obsolescence.
>>980515the thing with music is that there are already many many copyright strikes that will impact you IMMEDIATELY in music that analyzes your music and compares it to pre-existing songs and then gives that "owner" your money. Copyright is much more enforced in music than in images.
>>980517The music industry needs to be ripped out from the root. Restarted from scratch."The Big Three" music labels (UMG, Sony, WMG) have way too much fucking power over literally everything, even if it has fuck all to do with music. If AI sees it done, so be it. I'll still be listening to unsigned bands and shit anyway.I hope they're shitting in their beds. I mean they are, since it's run by old men who would be shitting in their beds due to incontinence, but I want the shit to be fear shit.
>>980807with music not only is copyright extremely heavily enforced to a degree that is not even close to other media but people want to see artists perform live in concert their own songs
>>980823>but people want to see artists perform live in concert their own songs>live
>>980825yeah, its a big thing. This normie i know just last week spent like over 3 hours driving somewhere to buy good tickets for a concert
>>980826you should tell the normie about the internet
>>980826>>980823I mean to be fair, I like to see the bands I like play live too. Gives you that reminder that they're real people (aside from Miku up there), and lets you hear their music in a different way than through a set of speakers, since a live show isn't something that's a perfect performance cut together in a recording studio.Of course, they only ever go to major population areas, and the nearest one is 3 hours away. I haven't bothered to make a show there because my car is a piece of shit and probably would die on the drive, but to me 3hrs isn't too bad an "investment" to go see a band I've listened to for 15 years or so. It's like driving a few hours to go to a theme park or something.
>>980825>comparing vocaloid to aiyou are a dumb nigger and have no clue what you're talking aboutai will always just be a sloppy soulless copy of something that would be a million times better made by human hands
>>980838Way to miss my whole point you colossal faggot
>>980511Current 3d ai peak is... this https://nsarafianos.github.io/garment3dgen You can now turn an image into a flat mesh with poor textures and baked in lighting that looks straight out of 2005
>>980511Ai to CG is going to take a whileImages can be seen as 100^2 while CG is 100^3 for example.Even direct AI video would probably be 'ready' before AI CG stuff.Not to mention that you could make slop, put it in a generator and have it shit out something better for stills
>>981038Not really because even when you're modeling by hand, you're still just creating a surface, even in 3D. It's just that the direct Crayola Magic Marker to AAA game AI pipeline has priority because Cris needs it. So, they're going to do that first.
the big nigger cums at midnight
>>981047What is that, Cris? Some hexing formula?
>>981059It's somekind of redditor handshake I think
There's this project that applies the ideas of noise/diffusion based image generation, like stable diffusion and combines it with gaussian splatting. The results are pretty good for what is essentially first gen. It even has browser based demo. https://gsgen3d.github.io/ if you can use this to generate a mesh, you could then use another program to retopo, and use image based ai to generate extra maps, or change texture styles.It wouldn't surprise me if some dev starts using this kind of workflow for background assets in the next 5 or so years. Especially with nanite allowing unnecessary details. Things like the fruits in a fruit stand in the next elder scrolls game, which is probably still run a gamebryo derivative...
AI will temporarily die the moment someone wins a trademark case. AI will have to be retrained with data that is not trademarked eventually
>>981250Not a single strike won because AI training on copyrighted data is legal and will remain legal. And what do you do when AI is trained on AI that is trained on copyrighted data? This shit should have sorted out in 2022 and the fact that it didnt fudge even an inch is sign that it will remain that way forever.
>>981894>And what do you do when AI is trained on AI that is trained on copyrighted data?ChatGPT and other AI networks already degraded because it's a cycle of shit, this doesn't work.
>>981895>>981894>>981250Honestly, "AI" (if you really can even remotely call it that) has peaked. There won't be any significant "game-changer" improvements in any capacity for a long time. Fundamentally, AI research the way it is now is pretty flawed and there's always going to be an upper limit of what's capable. Which we're pretty close to reaching. As they are now, all the computing power in the world won't make them smarter or more robust, just train them faster. Even training has diminishing returns, since training something too much ends up making it shit.AI is going to be stuck in "toyland" for a good while. Don't get me wrong, it's useful as it is, and there's certain applications that are really cool, but it'll never become a "real boy" like Pinocchio. The AI shit that seems like straight up magic won't happen for a really really long time. We'd need a significant breakthrough in computing and creating neural nets for something like that. Like we'd need a literal cyberbrain first.So yeah, in a sense "AI" is already dead. Or rather it hasn't even begun yet. What we have now is just a pale imitation of it by techbros. Neat and useful? Absolutely. But we're a long way off from something that'd really change the game enough to actually warrant fundamental policy change. Policy makers will probably do shit because they're old retards that don't know how tech works, but it's not really worth it at this stage. Image/video generation aren't something that'll shake the world to its core.
