NVIDIA’s Face Generator AI: This Is The Next Level! 👩‍🔬

  • Published on Jul 23, 2021
  • ❤️ Check out Fully Connected by Weights & Biases: wandb.me/papers
    📝 The paper "Alias-Free GAN" is available here:
    📝 Our material synthesis paper is available here: users.cg.tuwien.ac.at/zsolnai...
    🙏 We would like to thank our generous Patreon supporters who make Two Minute Papers possible:
    Aleksandr Mashrabov, Alex Haro, Andrew Melnychuk, Angelos Evripiotis, Benji Rabhan, Bryan Learn, Christian Ahlin, Eric Haddad, Eric Martel, Gordon Child, Ivo Galic, Jace O'Brien, Javier Bustamante, John Le, Jonas, Kenneth Davis, Klaus Busse, Lorin Atzberger, Lukas Biewald, Matthew Allen Fisher, Mark Oates, Michael Albrecht, Nikhil Velpanur, Owen Campbell-Moore, Owen Skarpness, Ramsey Elbasheer, Steef, Taras Bobrovytsky, Thomas Krcmar, Torsten Reil, Tybie Fitzhugh, Ueli Gallizzi.
    If you wish to appear here or pick up other perks, click here: www.patreon.com/TwoMinutePapers
    Or join us here: thexvid.com/user/keeroyz...
    Károly Zsolnai-Fehér's links:
    Instagram: twominutepa...
    Twitter: twominutepapers
    Web: cg.tuwien.ac.at/~zsolnai/

Comments • 1 685

  • ThioJoe
    ThioJoe  +953

    I feel like this is the exciting part where the low hanging fruit issues are mostly solved, and the really interesting & weird, previously-unimagined stuff can start being developed (for this specific facial generation purpose i mean).

  • AppSenence
    AppSenence  +570

    So in about 5 years from now we're gonna have our grandparents being called up on skype from someone who can make virtual videos of their loved ones and trick em to send money. Good game well played

  • anaisana
    anaisana  +252

    oh my god, now imagine them implementing a "Second camera Angle" to simulate how VR works, and then putting the images on a headset, now you have "fully 3d models" wild

  • Maxxx Modelz

    The implications of this possibility seem quite frightening to me. Allowing for really dubious practices and criminal activity in regards to framing people for crimes, identity theft, and much more. Unless the methodologies for uncovering AI-generated faces and videos are also being developed and advanced, this is likely going to fall into nefarious hands quickly.

  • Green Pumpkin

    The simulation of landscape looks like 3D slices of a 4D space. So moving along the 4th dimension and 3D objects just appears and disappears.

  • Code with Doga Ozgon

    Thank you for sharing this! NVIDIA really shows a good combination of what "attention to detail" and "exponential growth" looks like when they are put together.

  • Jorge C. M.

    This is the "two more papers down the line" hooray!

  • David E
    David E  +11

    The visualizations with animals and scenery are genuinely stunning

  • ActionScripter

    "But something is not right. Can you see it? Look closely."

  • Pɪɴɢᴜɪɴ Aɴɪᴍᴀᴛɪᴏɴs

    In Future, a kid will look at that and say, " well, that's not hard, i can do it in 3 seconds in my mobile".

  • alexandru benza

    The next time somebody asks me why I don't shave, I'm going to say, "the AI suffers from a sticky beard."

  • BenniK88

    Thank you for your beautiful presentation of that paper. I have a question. If it isn’t very computing heavy after it has been trained. How big in terms of data in mb is a final ‘real-time’ trained model like this? Is the final algorithm a lot of data? I guess so as describing faces seems quite complex.

  • Artemis Cane

    Wow, the representation of the internal processes looks a LOT like Alex Grey's art

  • chrisdotdash

    This is absolutely insane!! I love watching these videos. Nvidia is doing some crazy stuff!

  • 1qaz
    1qaz  +2

    I am truly blown away and I cannot imagine where this project will be 5 years from now!

  • MiPhinx
    MiPhinx  +9

    previous method still looks more convincing if you ignore the texture sticking. new method looks like a 2d picture being pulled on to emulate 3d.


    This truly is incredible. I fully understand now what you mean when you say “two papers down the line” because the difference from one paper to the next is usually only marginal, but it’s not until you compare it to the one after do you realise how much actually changes. At this rate it’ll only be another 4 papers until we’ll have ultra realistic video calls of (fake) people who’s mouth and facial expressions will match perfectly to natural sounding synthesised sound working as customer service operators. Get ready, because the future is oh so very nearly here, and it’s incredibly exciting 😄

  • Musonda Mapaalo

    This is insane!!! I can't wait to see this integrated into stock image libraries and graphic design software

  • panicdiver
    panicdiver  +161

    Those morphing videos were psychedelic. Amazing.

  • Pedro Braga

    I noticed myself the Texture Sticking, called it “Screenspace Textures”;