AI Image Generation

  • Guest, it's time once again for the massively important and exciting FoH Asshat Tournament!



    Go here and give us your nominations!
    Who's been the biggest Asshat in the last year? Give us your worst ones!
Tuco

Tuco

I got Tuco'd!
<Gold Donor>
47,358
80,734
Pasteton said:
Help me understand what is the roadblock from ai creating ai 3D worlds. Needs more data? Not enough compute ?
Click to expand...
You can probably find more educated answers via google, but one of the parts of this challenge is that a lot of AI right now operates on 2D images using some form of convolutional neural networks ( Convolutional neural network - Wikipedia ). Basically, we've found a lot of smart ways to take an image (which is just a bunch of numbers in a 2d array or matrix) and multiply them by stuff to get interesting outputs. This can be utilized in a huge variety of ways and a lot of times the way to build AI on non-image based systems is to first convert them into an image. Baidu's Apollo's prediction model works like this for example: https://arxiv.org/pdf/2006.06715.pdf

where information is encoded into a video-game like 2d top down map with colors, and then some 2d image networks process it.

The move from 2d art to 3d art (including 2d textures wrapped on a 3d polygon mesh and the 3d polygon mesh themselves) can't directly leverage those neural networks in the same way. Below is some image->3D stuff NVIDIA recently published:



research.nvidia.com

Neuralangelo

Project page for Neuralangelo: High-Fidelity Neural Surface Reconstruction.
research.nvidia.com research.nvidia.com

You can read more about there, but they process on a 3D hash map I don't really understand.

To answer your question directly: we have plenty of data/compute and the meganerds prioneering it just need time to develop the algorithms and technology to process it. Which is very exciting!
 
  • 1Like
Reactions: 1 user
Control

Control

Ahn'Qiraj Raider
2,983
7,878
While you might not be able to create full-3d the same way that Midjourney currently generates images, it's pretty easy (relatively speaking) to procedurally generate 3d spaces given a set of assets.
That's probably sufficient for most of the applications of 3d spaces, not to mention that there are probably orders of magnitude fewer potential users of 3d world ai than text or image ai, so far, far less demand too.
 
Raes

Raes

Vyemm Raider
3,264
2,720
Crude stuff, but I wonder what it will be like in a year or two.

 
  • 1Double Worf
  • 1Bullshit
  • 1Picard
Reactions: 2 users
pharmakos

pharmakos

soʞɐɯɹɐɥd
<Bronze Donator>
16,305
-2,234
USRyTvbWpq5VsO9Qa2ZJ--grid.jpg
NvpFyRkNEocnJzh9MVuH--8--aatd1.jpg
eiXcGFRx4uYxFIQoqw9Y--1--z752t_6.9444x.jpg
90RJjFh4nYKf7oHebaUf--1--pi15r.jpg
8uMuMaF3xaLCZlhjygg4--1--mz0a6.jpg
JVuZogB7cOaYULEcbAkp--1--chfx2.jpg
MTjz2MvlVl9cxQQHJEfM--1--2cr72.jpg
QS9GjxpJsxSx4tw38fSx--1--t0eyo.jpg
 
  • 1Like
Reactions: 1 user
Daidraco

Daidraco

Avatar of War Slayer
10,051
10,366
One of the girls I used to know moved away. She was kind of cute, not anything special. She put up a picture, and its got her facial structure, but something seems off. Im not trying to blow her spot up by just linking it here. But is it possible to put your face into one of these programs and it just completely alter your physique like hers? Her makeup is perfect etc. Cant find any tell-tale signs like disfigured hands or anything. Hell, if she can do that - fuck, are we in trouble.
 
Kharzette

Kharzette

Watcher of Overs
5,341
4,072
Yes there are AI filters, many that even work in realtime. Chinese girls use them alot for streaming.

There are some funny videos out there where the lady's filter malfunctions and she doesn't know, and you can see the viewer count plummet. :emoji_laughing:
 
pharmakos

pharmakos

soʞɐɯɹɐɥd
<Bronze Donator>
16,305
-2,234
Daidraco said:
One of the girls I used to know moved away. She was kind of cute, not anything special. She put up a picture, and its got her facial structure, but something seems off. Im not trying to blow her spot up by just linking it here. But is it possible to put your face into one of these programs and it just completely alter your physique like hers? Her makeup is perfect etc. Cant find any tell-tale signs like disfigured hands or anything. Hell, if she can do that - fuck, are we in trouble.
Click to expand...
You could train a model on nothing but pictures of yourself at your best made up, and generate countless new fake pictures of yourself. See things like the glut of Scarlett Johansson AI generated photos on social media etc.
 
Lambourne

Lambourne

Ahn'Qiraj Raider
2,862
6,829
Many phones even have these "beauty filters" as they're called built in now. Some social media even does a forced pass so users look better, get more attention as a result and end up using that platform even more.



What's worse, phone reviews tend to focus on picture quality so many phones now use a forced filter to make all pictures taken with them look better (more vibrant colours etc) so it's becoming harder to take pictures that haven't been doctored. Some Samsung phones detect the moon in pictures and replace it with a better looking version so the picture looks better. Not just upscale, outright replace.

www.wired.com

Samsung’s Moon Shots Force Us to Ask How Much AI Is Too Much

We like HDR-pimped pics, but this latest camera controversy outlines how computational photography could set us down a dicey path.
www.wired.com www.wired.com
 
  • 2Worf
  • 1Like
Reactions: 2 users
Tuco

Tuco

I got Tuco'd!
<Gold Donor>
47,358
80,734
Lambourne said:
www.wired.com

Samsung’s Moon Shots Force Us to Ask How Much AI Is Too Much

We like HDR-pimped pics, but this latest camera controversy outlines how computational photography could set us down a dicey path.
www.wired.com www.wired.com
Click to expand...
I like how Wired makes low quality regurgitation of articles about reddit posts, doesn't link the original reddit post (from what I can tell) and has the audacity to paywall it.



This cracks me up. I roll my eyes when people crack out their phones to take pics of moons or fireworks or whatever, but now with AI replacing their trash pics with HQ training data, maybe that won't be such a joke anymore.
 
  • 1Like
Reactions: 1 user
C

Captain Suave

Caesar si viveret, ad remum dareris.
5,253
8,953
tldw; DLSS 3.5 is giving major framerate increase (up to 3-5x) by generating 7 of every 8 pixels with neural AI instead of traditional rendering methods.

 
  • 1Like
Reactions: 1 user
You must log in or register to reply here.