AI Portraits | Hacker News Reader

AI Portraits

aiportraits.com

170 pointswillyyr53 comments

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

Why are they doing that?

Copy paste programming

Wow. Did they copy the same test for both fields twice too? Or more likely: do they have tests at all.

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).

I mean, not being able to sign up is fairly problematic.

True, but "server overwhelmed" is not usually one of the first tests written.

I'm referring to the inability to choose a password without some esoteric email-like rule-set.

I worked on a site supporting voter calls in 2004. There was no sanity checking of the usernames people proposed, so we had gems like leading spaces and multiple embedded spaces in the usernames. I figured some people signed up and then were never able to login.

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

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

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

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?

Just upload a picture of your dog. Pretty soon the tech will be good enough to identify you by your dog.

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.

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.

Friendly warning: people can see your face when you leave the house.

Yeah, but it's not like myriad cameras are recording their face and storing it for posterity whenever they leave the house! /s

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.

Let me tell you a story from about 10 years ago. I was working for the DOJ and was invited to one of their secure datacenter (I was cleared). As we are walking through hallways, we sometimes had to cross through rooms of racks of servers to get to other server rooms. As we walked through past these racks of servers, I noticed they all said EMC on the cages. My coworker points out each room we walk through is housing storage and measuring a single server room alone, you'd find they were pushing a few petabytes. Just imagine what their storage solution looks like now. Thats just the DOJ.

Another story.. 15 years ago, not many people had a digital camera. Almost overnight, everyone and their mom had a digital camera. My dog has a camera.

If you think your image is not captured and stored somewhere for a near indefinite period of time, then you are just fooling yourself. 1984 came and went and we didn't even notice we're living a brave new world now

I can simultaneously believe both you and dr_zoidberg, just as I can believe the homophobic totalitarian dictatorships of the Soviet Union were simultaneous with the 60s summer of love and Denmark decriminalising all pornography. They did say it “depends on where you live”.

While I think this is great for generally alleviating the creepiness of surveillance while keeping the benefits, why do they have such little storage? I'd have thought storing it would be the easy part compared to maintaining the whole systems.

In London they are. I assume you don't live there?

You can just try it with this: https://thispersondoesnotexist.com

What FaceApp incident?

There wasn't really an "incident".

Someone bothered to read the terms of service and noticed that they stated that you grant the app, the app's creator, and anyone else the app's creators chose, can do what-ever they like, when-ever they like, as often as they like, for as long as they like, with your photos and any other data they scrape from you as you use the app. This cause a bit of a stir, despite being pretty much what any similar ToS says.

And apparently the app creators are based in Russia which cased further concern to some.

I just uploaded a picture of nicolas cage.

> Didn’t the FaceApp incident _just_ happen?

There was no “incident”. Some people read the terms of service and raised a kerfuffle over it because the devs are based in Russia. So far we have no knowledge nor reasonable suspicions of anything nefarious on their part.

The "Who" tells you it's MIT folk; the confirmation message tells you the image will be immediately deleted.

The information (if you trust it) is clearly there on the website.

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.

Usually higher resolution will require significantly more GPU memory.

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-i...

It's down now.

getting the old HN hug-of-death atm

Well this was also on MIT's The Download. So I think more than just HN crashed it.

Also I saw it on The Verge

They should have used Mongo, I hear it's webscale.

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

The whole site seems to be down for me right now. EDIT: Just really slow. Seems to be overloaded.

Now its returning 404

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"

It's a javascript link. Try clicking rather than opening in new tab. Worked fine for me.

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

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

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

When hacker news traffic DDoS's your site

Got to much attention.

Site's down now...

JavaScript version?

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.

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.

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.

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.

> So it's 10 seconds for 200kb?

no, whole site took ~10 seconds to load with chromes Fast 3G throttled network preset.

Reply on news.ycombinator.com