Related: I know the authors used a DCGAN implementation (a pre-trained model, it looks like), but is it known what their approach for up-scaling the generated images is? In art generation I've seen GAN output of 128x128, e.g. that is then upscaled with a super-resolution network. Is something similar being done for the "final painting", or is the GAN somehow efficient enough to do large-format output in a decent trai…
Christie’s sells its first AI portrait for $432k
21–30 of 146 posts
Re: Christie’s sells its first AI portrait for $432k
#22Color me extremely unimpressed. I just can't believe someone would pay that much for something... so visually mediocre. When you look at how GANs work, they are incredible for what they are (especially StackGAN / GANs conditioned on some input)... but I just don't think there's any real creativity - or for that matter, intelligence - in GANs yet.
Re: Christie’s sells its first AI portrait for $432k
#23Related: I know the authors used a DCGAN implementation (a pre-trained model, it looks like), but is it known what their approach for up-scaling the generated images is? In art generation I've seen GAN output of 128x128, e.g. that is then upscaled with a super-resolution network. Is something similar being done for the "final painting", or is the GAN somehow efficient enough to do large-format output in a decent trai…
The short explanation is: Train and freeze the most basic layer of the model progressively to "let the network" understand higher resolution concepts one piece at the time, and avoid mode-collapse.
The network architecture illustrating this a bit better is shown on page 3 of the paper [1]
Re: Christie’s sells its first AI portrait for $432k
#24This could be something great if the author of the original algorithm and datasets was given proper credits. https://twitter.com/DrBeef_/status/1055360024548012033
I think that people don't seem to understand that most open source licences don't require attribution. If you have released your code according to a licence, what right do you have to complain that people are using that code in accordance with that licence? In this case though, it looks like Obvious did set out to attribute the original GAN author and Christie's have omitted that.
It’s not legally required to credit the author of the original code, but it’s a jerk move not to do so, especially if the work is primarily not your own.
All it does is discourage developers from releasing things open source, which is very bad in the long run.
Re: Christie’s sells its first AI portrait for $432k
#25Color me extremely unimpressed. I just can't believe someone would pay that much for something... so visually mediocre. When you look at how GANs work, they are incredible for what they are (especially StackGAN / GANs conditioned on some input)... but I just don't think there's any real creativity - or for that matter, intelligence - in GANs yet.
I have the same impression. It reminds me of the infamous “Ecce Homo” from Boria.
Re: Christie’s sells its first AI portrait for $432k
#26The results are so amateurish though. What’s with those nasty deconvolution artifacts?
Re: Christie’s sells its first AI portrait for $432k
#27Color me extremely unimpressed. I just can't believe someone would pay that much for something... so visually mediocre. When you look at how GANs work, they are incredible for what they are (especially StackGAN / GANs conditioned on some input)... but I just don't think there's any real creativity - or for that matter, intelligence - in GANs yet.
Re: Christie’s sells its first AI portrait for $432k
#28The next logical thing is to sell a computer and a large 4K monitor that creates a new painting for you every day
Re: Christie’s sells its first AI portrait for $432k
#29As someone who works in creative coding as an artist and has worked in advertising (where this happens a lot) I feel deeply sorry that someone made so much from this artists work they happily put out into the world, I’ve seen this happen time and time again. This is a really horrible life lesson that licenses matter and if you create something it’s important to think through yourself how to make money off it and do i…
I agree with you, except for the words "small bit". As an inventor, with a few issued and pending patents under my belt, I can assure you that bringing an invention (software or hardware) to market is no small feat.
For me, concieving of crafty algorithms or a new device is the easy part. Even writing a patent, whild time consuming, is straight forward. However, getting people to buy said inventions and establishing an efficient "supply chain", for me, is the excrutiatingly hard part. For some, this part comes naturally, and it's the act of invention that is tough. For a lucky select few, they are talented at both. Still, the requisite investment of time and money is generally not small.