14 ms·

AI Portraits

170 points by willyyr 7y ago

Waterluvian 7y ago
Does AI image generation struggle with high resolutions or is that just a choice for web friendliness? I was thinking maybe the limited resolution of these tools is to hide blemishes. I'd really like ultra high resolution versions of some of these.
- chrischen 7y ago
  Usually higher resolution will require significantly more GPU memory.
- yeldarb 7y ago
  Traditionally, GANs have. It wasn’t really possible to make high quality, high resolution GAN output until ~last year when researchers discovered how to progressively grow them from smaller resolution ones. https://towardsdatascience.com/progan-how-nvidia-generated-images-of-unprecedented-quality-51c98ec2cbd2 https://towardsdatascience.com/progan-how-nvidia-generated-i...
skilled 7y ago
Some of the pics I uploaded actually came out looking pretty rad. Would definitely find a use for them somewhere.
mkl 7y ago
The "How", "Why", and "Who" links aren't working for me. Are they working for anyone else?
- chrischen 7y ago
  The whole site seems to be down for me right now. EDIT: Just really slow. Seems to be overloaded.
  - oneinfiniteloop 7y ago
    Now its returning 404
- willyyr 7y ago
  It loads but very slow. Most of them seem to be from the MIT IBM Watson AI Lab. I like that they used generated portraits of themself :D Extract of "Why": "AI Portraits Ars uses Artificial Intelligence to reproduce artistic human portraits, with different styles and levels of abstraction. For our model training, we adopt a data set of tens of thousands of paintings from the Early Renaissance to Contemporary Art. This type of portraiture is quite distinctive of the Western artistic tradition. Training our models on a data set with such strong bias leads us to reflect on the importance of AI fairness. In the previous work AI Portraits Celebrity, we explored the concept of micro-bias linked to the training data of only actors, which in some way imposes an actorization of the user's portrait: “a collection of faces from the society of spectacle that are sedimented in the neural network, and vaporize my selfie in a cinematographic self." AI Portraits Ars introduces a very different type of bias with unique themes to explore." // edit: Added extract of "Why"
- beilabs 7y ago
  It's a javascript link. Try clicking rather than opening in new tab. Worked fine for me.
probst 7y ago
In case you want to sign-up and cannot get the client side validation to accept your password: they apply email validation to the passwords. You'll have to make up a password that's also a valid email address. In other words, something like: aBqwdHkjh@oqwIrd.com
- metahost 7y ago
  Why are they doing that?
  - dvh 7y ago
    Copy paste programming
- cameronbrown 7y ago
  Wow. Did they copy the same test for both fields twice too? Or more likely: do they have tests at all.
  - cpfohl 7y ago
    Lots of places don't have tests at all. This is not uncommon. (Or even all that problematic if other processes and capabilities are in place).
    - cameronbrown 7y ago
      I mean, not being able to sign up is fairly problematic.
      - pixelbath 7y ago
        True, but "server overwhelmed" is not usually one of the first tests written.
        cameronbrown 7y ago
V-2 7y ago
It's down now.
- CodeBiscuit 7y ago
  getting the old HN hug-of-death atm
  - CodiePetersen 7y ago
    Well this was also on MIT's The Download. So I think more than just HN crashed it.
    - Rerarom 7y ago
      Also I saw it on The Verge
- russh 7y ago
  They should have used Mongo, I hear it's webscale.
tarcyanm 7y ago
The wispy, near transparent spectacles in some of them are a bit of a give away...
m4r 7y ago
Open dev tools and select networking - this page performs so many concurrent requests to load images, it will never work on medicore connection using mobile device. It even fetches/tries to renders one that are 404.
- shusson 7y ago
  It's not that bad. The images are ~20kb each and it fetches ~10 images on the landing page. If I throttle the network to Fast 3G in chrome, the whole site takes about 10 seconds to load.
  - m4r 7y ago
    So it's 10 seconds for 200kb? From my perspective its unacceptable, if your service is working with images, at least present them properly. Also, it loads more than 10 when you start scrolling.
    - dymk 7y ago
      Why does every thread have to have the same pithy, cheap, criticizing comment about page load speed? It’s demoing a GAN, not a shiny “blazing fast” web framework. They whipped this web frontend up in a day probably. It’s not supposed to be optimized for your 3G phone.
    - shusson 7y ago
      > So it's 10 seconds for 200kb? no, whole site took ~10 seconds to load with chromes Fast 3G throttled network preset.
DanielleMolloy 7y ago
Attach this to one of these Chinese oil painting factories: https://www.artsy.net/article/artsy-editorial-village-60-worlds-paintings-future-jeopardy https://www.artsy.net/article/artsy-editorial-village-60-wor...
- chrischen 7y ago
  Many of the Chinese factories are available to order from on https://www.Instapainting.com https://www.Instapainting.com (I'm the founder).
nkcmr 7y ago
So, I should upload a picture of my face to some random website/app to let it do something silly with my face and just let them have a picture of my face? Didn’t the FaceApp incident _just_ happen? What’s the privacy policy for this website?
- not_a_cop75 7y ago
  Just upload a picture of your dog. Pretty soon the tech will be good enough to identify you by your dog.
- adrianolek 7y ago
  The "how" link in the top-right corner says: Your photos are sent to our servers to generate portraits. We won’t use data from your photos for any other purpose and we'll immediately delete them. So basically you have to trust them.
  - giarc 7y ago
    But then the page has a ton of photos (presumably submitted by users) that I can download. So perhaps the original photo isn't used and deleted, but the generated photo belongs to them.
- nmeofthestate 7y ago
  Friendly warning: people can see your face when you leave the house.
  - stronglikedan 7y ago
    Yeah, but it's not like myriad cameras are recording their face and storing it for posterity whenever they leave the house! /s
    - dr_zoidberg 7y ago
      Depends on where you live. Where I live, I know for a fact they have less than 1TB of storage for 1000 cameras, and law enforcement usually fights the city to get images of incidents, because they don't live over a day due to storage shortage.
leanthonyrn 7y ago
JavaScript version?
mellosouls 7y ago
They are clearly having traffic issues at the mo, maybe HN and other related raised visibility.
Balgair 7y ago
Combine with https://thispersondoesnotexist.com/ https://thispersondoesnotexist.com/ to have some fun, without having to upload a 'real' photo
runxel 7y ago
Got to much attention. Site's down now...
CTrom 7y ago
When hacker news traffic DDoS's your site
satanas2 7y ago
It's fucking down now! What happened?

170 points by willyyr 7y ago

Does AI image generation struggle with high resolutions or is that just a choice for web friendliness? I was thinking maybe the limited resolution of these tools is to hide blemishes. I'd really like ultra high resolution versions of some of these.

chrischen 7y ago
Usually higher resolution will require significantly more GPU memory.
yeldarb 7y ago
Traditionally, GANs have. It wasn’t really possible to make high quality, high resolution GAN output until ~last year when researchers discovered how to progressively grow them from smaller resolution ones. https://towardsdatascience.com/progan-how-nvidia-generated-images-of-unprecedented-quality-51c98ec2cbd2 https://towardsdatascience.com/progan-how-nvidia-generated-i...

skilled 7y ago

Some of the pics I uploaded actually came out looking pretty rad. Would definitely find a use for them somewhere.

mkl 7y ago

The "How", "Why", and "Who" links aren't working for me. Are they working for anyone else?

chrischen 7y ago
The whole site seems to be down for me right now. EDIT: Just really slow. Seems to be overloaded.
- oneinfiniteloop 7y ago
  Now its returning 404
willyyr 7y ago
It loads but very slow. Most of them seem to be from the MIT IBM Watson AI Lab. I like that they used generated portraits of themself :D Extract of "Why": "AI Portraits Ars uses Artificial Intelligence to reproduce artistic human portraits, with different styles and levels of abstraction. For our model training, we adopt a data set of tens of thousands of paintings from the Early Renaissance to Contemporary Art. This type of portraiture is quite distinctive of the Western artistic tradition. Training our models on a data set with such strong bias leads us to reflect on the importance of AI fairness. In the previous work AI Portraits Celebrity, we explored the concept of micro-bias linked to the training data of only actors, which in some way imposes an actorization of the user's portrait: “a collection of faces from the society of spectacle that are sedimented in the neural network, and vaporize my selfie in a cinematographic self." AI Portraits Ars introduces a very different type of bias with unique themes to explore." // edit: Added extract of "Why"
beilabs 7y ago
It's a javascript link. Try clicking rather than opening in new tab. Worked fine for me.

probst 7y ago

In case you want to sign-up and cannot get the client side validation to accept your password: they apply email validation to the passwords. You'll have to make up a password that's also a valid email address. In other words, something like: aBqwdHkjh@oqwIrd.com

metahost 7y ago
Why are they doing that?
- dvh 7y ago
  Copy paste programming
cameronbrown 7y ago
Wow. Did they copy the same test for both fields twice too? Or more likely: do they have tests at all.
- cpfohl 7y ago
  Lots of places don't have tests at all. This is not uncommon. (Or even all that problematic if other processes and capabilities are in place).
  - cameronbrown 7y ago
    I mean, not being able to sign up is fairly problematic.
    - pixelbath 7y ago
      True, but "server overwhelmed" is not usually one of the first tests written.
      - cameronbrown 7y ago

V-2 7y ago

It's down now.

CodeBiscuit 7y ago
getting the old HN hug-of-death atm
- CodiePetersen 7y ago
  Well this was also on MIT's The Download. So I think more than just HN crashed it.
  - Rerarom 7y ago
    Also I saw it on The Verge
russh 7y ago
They should have used Mongo, I hear it's webscale.

tarcyanm 7y ago

The wispy, near transparent spectacles in some of them are a bit of a give away...

m4r 7y ago

Open dev tools and select networking - this page performs so many concurrent requests to load images, it will never work on medicore connection using mobile device. It even fetches/tries to renders one that are 404.

shusson 7y ago
It's not that bad. The images are ~20kb each and it fetches ~10 images on the landing page. If I throttle the network to Fast 3G in chrome, the whole site takes about 10 seconds to load.
- m4r 7y ago
  So it's 10 seconds for 200kb? From my perspective its unacceptable, if your service is working with images, at least present them properly. Also, it loads more than 10 when you start scrolling.
  - dymk 7y ago
    Why does every thread have to have the same pithy, cheap, criticizing comment about page load speed? It’s demoing a GAN, not a shiny “blazing fast” web framework. They whipped this web frontend up in a day probably. It’s not supposed to be optimized for your 3G phone.
  - shusson 7y ago
    > So it's 10 seconds for 200kb? no, whole site took ~10 seconds to load with chromes Fast 3G throttled network preset.

DanielleMolloy 7y ago

Attach this to one of these Chinese oil painting factories: https://www.artsy.net/article/artsy-editorial-village-60-worlds-paintings-future-jeopardy https://www.artsy.net/article/artsy-editorial-village-60-wor...

chrischen 7y ago
Many of the Chinese factories are available to order from on https://www.Instapainting.com https://www.Instapainting.com (I'm the founder).

nkcmr 7y ago

So, I should upload a picture of my face to some random website/app to let it do something silly with my face and just let them have a picture of my face? Didn’t the FaceApp incident _just_ happen? What’s the privacy policy for this website?

not_a_cop75 7y ago
Just upload a picture of your dog. Pretty soon the tech will be good enough to identify you by your dog.
adrianolek 7y ago
The "how" link in the top-right corner says: Your photos are sent to our servers to generate portraits. We won’t use data from your photos for any other purpose and we'll immediately delete them. So basically you have to trust them.
- giarc 7y ago
  But then the page has a ton of photos (presumably submitted by users) that I can download. So perhaps the original photo isn't used and deleted, but the generated photo belongs to them.
nmeofthestate 7y ago
Friendly warning: people can see your face when you leave the house.
- stronglikedan 7y ago
  Yeah, but it's not like myriad cameras are recording their face and storing it for posterity whenever they leave the house! /s
  - dr_zoidberg 7y ago
    Depends on where you live. Where I live, I know for a fact they have less than 1TB of storage for 1000 cameras, and law enforcement usually fights the city to get images of incidents, because they don't live over a day due to storage shortage.

leanthonyrn 7y ago

JavaScript version?

mellosouls 7y ago

They are clearly having traffic issues at the mo, maybe HN and other related raised visibility.

Balgair 7y ago

Combine with https://thispersondoesnotexist.com/ https://thispersondoesnotexist.com/ to have some fun, without having to upload a 'real' photo

runxel 7y ago

Got to much attention. Site's down now...

CTrom 7y ago

When hacker news traffic DDoS's your site

satanas2 7y ago

It's fucking down now! What happened?