NVIDIA’s New AI Grows Stuff Out Of Nothing!

❤️ Try Macro for free and supercharge your learning:

📝 The paper "Meshtron: High-Fidelity, Artist-Like 3D Mesh Generation at Scale" 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:
Benji Rabhan, B Shang, Christian Ahlin, Gordon Child, John Le, Juan Benet, Kyle Davis, Loyal Alchemist, Lukas Biewald, Michael Tedder, Owen Skarpness, Richard Sundvall, Steef, Taras Bobrovytsky, Thomas Krcmar, Tybie Fitzhugh, Ueli GallizziIf you wish to appear here or pick up other perks, click here:

My research:
X/Twitter:
Thumbnail design: Felícia Zsolnai-Fehér –

#nvidia

Joe Lilli
 

  • @TheAkdzyn says:

    Thanks for the insights. I can only imagine what things will look like two papers down the line. This technology would be excellent for prototyping!

    • @yeenybean2881 says:

      @@TheAkdzyn Let’s not lie. This won’t be used for prototyping. It will completely replace artists. Claiming otherwise is simply illogical.

    • @christophkogler6220 says:

      ​@@yeenybean2881 3D modelers usually do a LOT more than just converting objects to vertices and topology. So far, only texturing and model creation have been (kind of, mostly) solved, and the job is still pretty safe. You can probably make an automated chain to go text -> point cloud -> model -> textured model already, but it would only be a prototype. It would absolutely need significant cleanup and refinement, especially if you want a cohesive style across multiple creations. And that’s still missing the immense amount of effort that goes into producing high quality rigs.

      It is likely that in the coming years the entire chain will be automated to near-human level. But, by that point, so will most digital work, and recent advancements in robots will have likely automated most simple physical labor, so the entire concept of human mental or physical labor having monetary value faces serious conceptual problems. Any artists will be working for their own joy of creation, not producing goods for money.

    • @zackhack_gamer2487 says:

      Isn’t that the most fun part of the process…? You imagine something and get to create it.. Decide how it looks, find out about how else it can be done..
      At this point you’re simply replacing yourself with ai.

      You’re speaking as if creativity and fun isn’t the point, but money. Shame on you.

    • @dude3278 says:

      @@yeenybean2881 That’s totally fine with me

    • @dude3278 says:

      ​@@yeenybean2881im totally fine with that

  • @luciosergiocatilina1 says:

    What a time to lose your job!

    • @-Lambda says:

      What a time to lose all potential to do something creative that you enjoy and actually be able to make a living off of it!

    • @IceMetalPunk says:

      @@-Lambda Sounds like a problem with capitalism to me.

    • @WDShorty says:

      Remember that anything created with AI can’t be copywrite protected

    • @begobolehsjwjangan2359 says:

      nope.
      virtual designers would work faster because client would come to them with already meshed model and ask them to give retouch or whatever the details they want.

    • @contabilidade00imperiocont95 says:

      @@-Lambda so true. I am a programmer and am so glad AI is here. So much tedious work I don’t have to do anymore and making staff I really care about.

  • @pstefan86 says:

    Slightly shocked to see myself at 6:11 squeezing that paper 😂

  • @toop4538 says:

    we got 2d artists sweating, now we about to get 3d artists drenched in sweat.

  • @ChaosSwissroIl says:

    We are getting closer to plain text generation of G-code. Once we get that, the consumer manufacturing revolution is inevitable.

    • @fischX says:

      We are really close to that it’s basically plumbing together existing tools – at least for 3D printing g-code that is. To get your individual prompt based Ferrari with a duck face manufactured in front of you ready for a test drive in 10 minutes will be take a while – but we will get there

    • @ChaosSwissroIl says:

      @@fischX I think firearms will lead the way because they’re high value, custom works are very expensive, the most desirable examples are highly finite, and they’re small and therefore the machinery to produce them is low cost and low footprint.

    • @jackdusch8896 says:

      I’m working on an app for this right now. If you or anyone is reading this I have some older videos and development is ongoing, sometimes I stream development and I’ll be testing live soon

    • @andrewdavie386 says:

      @@ChaosSwissroIl in the USA, maybe, where firearms are a thing. Not so much in the rest of the world – i.e., 95% of the people on the planet.

  • @mikesbasement6954 says:

    And yet they still cannot make an affordable video card.

  • @MikkoRantalainen says:

    This looks really promising! This seems to be already at the level where I would want to have it for retopology in Blender – there the “input point cloud” could be full sculpted mesh and the output would be the same mesh with improved topology. Even if it didn’t do perfect job but mostly good, it would speed up retopology workflows a lot.

  • @teamredstudio7012 says:

    This is just AMAZING!! We can generate point clouds too using our phone! iPhones have had Face ID for 8 years and some also have LiDAR which can both be used to make accurate point clouds. This means we can use our phones to generate actually high quality models of real life objects! In the future you just scan your face with FaceID and you are accurately modelled and put in game!

    • @teamredstudio7012 says:

      Combined with speech to text, an LLM and text to speech, you can clone real humans, in 3D, with their own voice and face into a virtual world. Let’s see what we can do, 20 more papers down the line. Incredible! What a time to be alive!

    • @assarlannerborn9342 says:

      @@teamredstudio7012 soon my love soon i will see you my future 3d wife

    • @acertainredburnygirl5265 says:

      @@teamredstudio701225 paper down and we’ll be able to fully digitalize our brain! We can create an ideal reality and put our brain in there for eternal bliss, leaving the physical body all together… wait that sounds familiar

  • @ccchan2649 says:

    Huh, he didn’t say “What a time to be alive!!”. I feel cheated.

  • @phantomxt4427 says:

    optimised UV mapping and UV painting should be here any minute now.
    Looking forward to how ai can handle rigging assets and animation, maybe rigging is something that should be worked into the point cloud phase.

  • @contabilidade00imperiocont95 says:

    Wait!!! no “what a time to be alive”?

  • @mnhgnm says:

    Correcting topology is amazing

  • @thepr0m3th3an says:

    I love how AI continues to take the creative work away from humans but leaves the drudgery behind for us to complete. Great future we are heading towards… lol

  • @biruuzumaki4126 says:

    Why only cover papers about Nvidia we want see more diverse papers

  • @deletedpvp4703 says:

    what a time to be an AI

  • @ibie27 says:

    Combining this with the Quad Remesher addon would be an absolute disaster for 3D modellors
    There goes my three years of 3D modelling 😅

  • @pfannkuchengesicht42 says:

    These days this channel seems to be just about various AI models from nvidia, Google, OpenAI and Microsoft. More diverse topics like a couple years ago would be appreciated.

  • @kanetsb says:

    Finally, they came for the 3D artists. 😉

    • @kyberite says:

      first we carried sh!t by hand, then the wheel came for the carriers. Then the plane came for the drivers. It’s a process, the end result is to eliminate work alltogether, unless you like this neo-feudal bs system that caters only to investors and literal human scum that’s only interested in profit

  • @carlosrivadulla8903 says:

    It never ceases to amaze me how they joyfully celebrate taking art away from human beings.

  • @xXDeiviDXx says:

    “This AI can generate models with high polygonal counts!”
    Shows a Cybertruck

  • >