Need Profile Photos? Generate Faces that never existed.
 

This is pretty technical, however maybe some of you will get some use out of it.

Using neural networks and training them to create original faces of people who never existed is entirely possible.

The following articles explain some of the different approaches.


https://medium.com/coinmonks/celebri...n-eaa2001eef86

https://blog.insightdatascience.com/...i-d170b1b59255

https://www.theverge.com/2018/12/17/...-networks-gans

Some code can be found here:

https://github.com/HACKERSHUBH/Face-...sarial-Network

https://github.com/akanimax/T2F

And some reference material:

https://medium.com/deep-dimension/ga...e-83ed64b42f5c

https://www.oreilly.com/learning/gen...-for-beginners

https://towardsdatascience.com/under...s-4dafc963f2ef

Looks interesting. Thank you for the links.

Need profile pictures?

Buy some content.

Generating original ones with a Neural Network is more fun!

Nobody can reverse image search a profile picture that can not possibly have existed before.

Need jail time for fraud? Then use it to open a dating site.

:1orglaugh

https://uifaces.co/

real ppl, real faces, volunteered .

:1orglaugh:thumbsup

i am gonna use this for my new dating site.

hold up, are you telling me that the ads on tubes letting me know there are thousands of women in my city (of less than 800k in population total) waiting to fuck me for free is a lie?

I feel very disillusioned :/

incidentally, is there a neural network that exists to make me look like Liya Kebede when I walk down the street? that would be awesome.

Interesting, could also be useful for getting around FB's "upload a picture of your face" roadblock, unless there's some hidden data in there giving it's origins.

if you are worried/paranoid then just dont use facebook, i do the same.

Pretty soon we won't need actors and actresses. It will all be computer generated.

Correct, eventually this kind of stuff will put a lot of stock photography models out of work.

More interesting though is the use of the underlying technology (generative adversarial networks) to replicate various types of things from music to faces to theoretical mutating diseases that may be cured before they come into reality to cancer research.

The scary thing is that these generative adversarial networks are being highly adopted in areas that we don't fully understand yet and research has shown they can amplify biases or prejudice in both photographic and written material they are trained on, take that a few steps further and we better hope that if this shit becomes sentient one day that it doesn't get the ability to hurt us.

Nice, would be good to have software which would do that automated.

Even on very fast GPU systems GANs take a long time to train and then get results from.

Well i suppose we should wait till they make it fast as at the moment seems it will be less hassle to play with photoshop instead.

If you put it that way then let me counter by saying it's quicker to create 1 million brand new faces that never ever existed with a GAN than any other method, including Photoshop. It would even be quicker to make 1000 with a GAN than 1000 in photoshop.

Yeah, but even the surfers know those are lies so they don't count.

:1orglaugh

Interesting Geoffrey!
 

Intriguing my man!!

:thumbsup:thumbsup

So these pics are public domain...since the people don't exist and can only be created by a page refresh from the page visitor.

You can teach a computer how to make a face...but can you teach it how to act? I think you would have to put some shoes on the computer and let it go live life for a while before it can do what Glenn Close does.