NVIDIA’s New AI: Game Changer!
❤️ Check out Lambda here and sign up for their GPU Cloud:
📝 The paper "XCube: Large-Scale 3D Generative Modeling using Sparse Voxel Hierarchies" is available here:
📝 My paper on simulations that look almost like reality is available for free here:
Or this is the orig. Nature Physics link with clickable citations:
🙏 We would like to thank our generous Patreon supporters who make Two Minute Papers possible:
Alex Balfanz, Alex Haro, B Shang, Benji Rabhan, Gaston Ingaramo, Gordon Child, John Le, Kyle Davis, Loyal Alchemist, Lukas Biewald, Martin, Michael Albrecht, Michael Tedder, Owen Skarpness, Richard Sundvall, Taras Bobrovytsky, Ted Johnson, Thomas Krcmar, Tybie Fitzhugh, Ueli Gallizzi.
If you wish to appear here or pick up other perks, click here:
My research:
X/Twitter:
Thumbnail design: Felícia Zsolnai-Fehér –
#nvidia
What a time to be alive!
I hear it somewhere 🤔🤔
Best drake album
bot
What a time to [This post has been redacted due to Google’s community standards.]
Holding on to my paper!
Awesome! Would love to see an AI model that does all the boring parts of 3D modeling like UV unwrapping and weight painting
so long as the AI model doesn’t get bored… yeah, would be good….
Been saying this for ages now. We need an AI tool that automated the UV unwrapping part in the most logical and unstretched way possible that makes it very clean and easy to paint on or repeat tiled textures on so that islands line up with the main body to prevent misalignment of the tiling texture
@@MonsterJuiced Many of you don’t realize there is a big reason why generative AI today is automating the fun creative parts of art, rather than the boring parts. It’s not out of a desire to automate fun away, but it’s just a technical inevitability, because of the nature of training data.
As we know, neural nets are unstoppable as long as they have massive amounts of training data. So, what kind of training data is easier to acquire in massive amounts: Labeling a bunch of raw models with UV data, or labeling a bunch of 3d models with words?
We already saw this happen in music. Since I was a kid, I had hoped for an AI that can transcribe music into notes or turn notes into music. But what did we get? An AI that goes straight from text to music, without even allowing you to specify or edit the notes and chords. That’s because it’s much easier to get training data of people labeling a piece as “melancholy violin piano” than to annotate all the actual notes.
And yes, training data does exist for those “harder” tasks, but in nowhere near as massive quantities.
@@MonsterJuiced There is a reason why generative AI today is automating the fun creative parts of art, rather than the boring parts. It’s not out of a desire to automate fun away, but it’s just a technical inevitability, because of the nature of training data.
As we know, neural nets are unstoppable as long as they have massive amounts of training data. So, what kind of training data is easier to acquire in massive amounts: Labeling a bunch of raw models with UV data, or labeling a bunch of 3d models with words?
We already saw this happen in music. Since I was a kid, I had hoped for an AI that can transcribe music into notes or turn notes into music. But what did we get? An AI that goes straight from text to music, without even allowing you to specify or edit the notes and chords. That’s because it’s much easier to get training data of people labeling a piece as “melancholy violin piano” than to annotate all the actual notes.
@@MaxLoh2 you don’t need an AI to turn notes into music, just a DAW
The 3D text generation won’t be good until the AI can automatically retopologize the geometry. Right now, there are way too many tris, and the models need intense retopology to become usable. Until then, I’ll keep an eye on it
be interested to know your thoughts on rodin, it doesn’t look like it uses the voxel method that a lot of text-to-ai uses that makes those lumpy models, but i haven’t used it for anything serious so i don’t know how well it does, but i had recently seen a video on youtube of a guy doing comparisons of different text-to-3d ai for 3d printing and rodin came out the clear winner.
But what about the topology?
been hoping for ages someone would make a system to retopologize and do quality uv unwrapping. I know i’m not the only artist that wants that. which ever company does listen to what we want, i feel like they’ll do well.
@@jameshughes3014 3DCoat does this pretty solid
Oh wow we’re on the edge of generative videogames, we’ll remember procedural games like we remember pong
No, we remember Pong because it was unique. Here we’ll probably get thousands of games that no one will play. Or, more likely, thousands of bad games that no one will play.
@@Rusu421 Nah, we’ll eventually get billions of games that one or two people will play because the game was generated and personalized for them.
This will make CAD and 3Printing so much easyer in the near future…
I was an AI branch student in 2019, but I quit college to pursue my passion for art. I was doing decently well until AI started gaining traction, and I realized I needed to take the initiative before my commissions dried up. So, I learned 2D animation. But as AI began making strides in 2D animation too, I knew I had to stay ahead of the curve, so I started learning 3D modeling.
Today, I’m an intermediate 3D modeler, in the process of building a portfolio. Unfortunately, all my 2D commissions have completely dried up, and now I’m seeing AI start to make inroads into 3D as well.
I’m an optimistic person and believe that I’ll be fine on an individual level. I understand that being unlucky and needing to innovate is part of life. However, on a systemic level, I hope that our society finds a way to support people like me.
I’m an expert-level user at 2d and 3d, not so much animation, but more still images. I have over 40 years experience of both analog and digital illustration and sculpting. Like you I’ve done my utmost through the years to stay ahead of technical innovation.
But now – imo unless new laws are created making generative AI illegal – either the corporations themselves, or the models they’re currently selling, or the commercial use of AI, or something in that vein – all of us artists (including all so called ‘prompt-artists’) will be looking for entry level jobs in other sectors of the economy soon.
@@stahlbergpatreon6062 this comment was made “2 minutes ago” nice, But other than that, This AI is pretty expensive to run , The electricity, the CO2 emitted just to “generate” “Ai” “Art?”
every failed attempt may be forgotten but it still takes resources to do
Microsoft (i think) building a new AI Data center (or server?) That ll consume enough electricity to power a small country (like WTF) also sry for bad english
@@stahlbergpatreon6062 any advise what those roles could be?
Would love your insights
@@Faizan29353as they say, this current AI is the worst it will ever be; it will only get better and more efficient in the future. And if it doesn’t, I don’t think they’ll care about the CO2 cost. They’ll likely use the race to AGI with China as a reason to ignore the risks.
@@Faizan29353 as they say, this current AI is the worst it will ever be; it will only get better and more efficient in the future. And if it doesn’t, I don’t think they’ll care about the CO2 cost. They’ll likely use the race to AGI with China as a reason to ignore the risks.
having AI be able to navigate better will be great for regular robots too, not just cars
this is why i was so excited about metas recent segment anything work. We’re getting closer to having a Rosie the robot to help with laundry and chores, i love it.
It’s crazy that even two minute papers seems behind on the latest AI. Mew things are happening daily 😮 What a time to be alive!
One step closer to getting the oasis of ready player one in real life.
The models right now resemble the object very well, but are still auite uncanny. I’m sure people will definitly be able to use this as a baseline though to skip a lot of the modelling process.
At some point we get “Text to triple A game”
You just type one sentence of description, and the AI is generating a fully playable experience for you.
as an indie game developer.. i wouldn’t even be mad lol. making games is HARD. even if we could generate a game with a prompt, we’d still need creatives to guide the machine to make it good in tons of ways. That would kinda be awesome.
This is coming along nicely. Later on all 3D Design Engines like Unreal, Houdini, Blender etc. will have midjourney like ai models that will help the artists speed things up insanely fast. Just ask for several burnt destroyed car models and done, drag and drop them into the scene. Several different models of trees and bam, drag and drop.
Unfortunately, this is still not very useful for game development, where you need to provide low / medium polygon models with a VERY CLEAN and optimized geometry. Detail richness is mostly created by texture layers, not with 3d geometry. These relatively ‘sloppy’, much too high-resolution 3d meshes as currently produced by AI, are of not much help here …yet 😐
Do you have stocks in nvidia? If yes, why?
I love the weird stuff. I feel like 3d generators are at that great place right now that image generators were in a couple years ago, where they can’t help but make things weird and surreal. I know that’s not the end goal but it’s my favorite.
“What will you fellow scholars use this for?”
Catgirls probably 🙂
4:25 is unbelievable. As if we’re alive when ai can make something that looks better than the first Toy Story movie. Wiiiild
I would love to see more assistive AI then genitive AI