All this talk about neural these days...Doctors say nerves don't regenerate... But that's not true... I have restored function to nerves in my body. Using neural networks also reminds me of the science fiction shows...Is it Star Trek that had the bad guys linked neurally? I prefer to keep true to my artistic individuality.
I love the neural net stuff. And have been diving into it very very deeply.
But at the same time, i'm becoming very aware as i look at what different artists are doing with it, that sometimes you look at the output and say, oh that looks cool, but then when you begin to analyze it you realize that you could have used an image folder brush in Studio Artist (or some other Studio Artist feature) to get there.
Instead, the artist is mucking around in colab using research PyTorch or Keras (tensorflow) code, running it on the cloud in a colab notebook, waiting hours and hours to intense gpu processing in the cloud, and then they get a 256x256 or maybe a 512x512 output art image run out through the neural net research code.
Research code they might be circuit bending in really interesting ways that never occurred to the neural net researchers who wrote it. So we are of course paying extra special attention to how they are repurposing this research code. Because we want to provide a better way to access that style or type of the effect. That is hopefully more easily manipulatable, infinitely more adjustable, and that scales to more useful resolutions as a default.
Here's the thing, ultimately all of the neural net stuff is just about analyzing statistics from sets of images, and then manipulating, or perhaps generating new unique images, based off of the aggregate statistical properties of the database of images used for training the neural net.
Neural nets are universal function approximators, so that is why they end up being great black boxes to train on real world data (images in the case of the things visual artists are interested in). they can adaptively learn to model the data by just being exposed to , and trained, on that data. Data again being collections of images when you are talking about what everyone who reads this users forum is interested in.
Neural nets are designed to loosely model what is going on with collections of neurons in the brain. Hence the term neural.
Ultimately it's just another tool for the artist. One that is a hot fast moving research topic. So people get excited, the press gets excited, one can develop a certain amount of 'neural net fetish' if your work is associated with it. But at the end of the day we're all trying to create visually compelling images. How we got there is less important than where there is, or maybe what there looks like.
Your last paragraph says it all. Computers have allowed us to analyze data in a way previously not imagined. However to my way of thinking there has to be a way to transcend the data for creativity to manifest itself. The deeper one goes into and becomes conscious of awareness within one owns brain and self, the greater the creative response will be. Intuition is not just a silly word. Why be a prisoner to data output? Try the road less traveled.
You could use something like the Vectorizer to transform a natural image of people into something more stylized like the source used for the processing in the image above. And then load that Vectorizer processed stylized image of the people as the new source, and then run the vector paint presets on it to get the painted look for it.
Here's an example from my art blog of what i mean by vectorizer processing a natural source image into something more stylized. The source image is a totally normal photo of 4 people.
Don't ask me for the Vectorizer preset, because it was generated on the fly by gallery show in Studio Artist V5.5 during automated testing, so it's totally a generative art thing designed by the AI generative system.
This reply was deleted.
You need to be a member of Studio Artist to add comments!
Comments
would you please upload the source image and style too.
Bernard, This is great!
I can see you have been working quietly, away from the maddening crowd, in a pasta factory!
Hahaha, perhaps looks like neural, but isn't ! Just an old vector/paint preset, and some other operations...
Here's the source :
Thanks Paul, yes, pasta neural network
Bernard,
All this talk about neural these days...Doctors say nerves don't regenerate... But that's not true... I have restored function to nerves in my body. Using neural networks also reminds me of the science fiction shows...Is it Star Trek that had the bad guys linked neurally? I prefer to keep true to my artistic individuality.
I love the neural net stuff. And have been diving into it very very deeply.
But at the same time, i'm becoming very aware as i look at what different artists are doing with it, that sometimes you look at the output and say, oh that looks cool, but then when you begin to analyze it you realize that you could have used an image folder brush in Studio Artist (or some other Studio Artist feature) to get there.
Instead, the artist is mucking around in colab using research PyTorch or Keras (tensorflow) code, running it on the cloud in a colab notebook, waiting hours and hours to intense gpu processing in the cloud, and then they get a 256x256 or maybe a 512x512 output art image run out through the neural net research code.
Research code they might be circuit bending in really interesting ways that never occurred to the neural net researchers who wrote it. So we are of course paying extra special attention to how they are repurposing this research code. Because we want to provide a better way to access that style or type of the effect. That is hopefully more easily manipulatable, infinitely more adjustable, and that scales to more useful resolutions as a default.
Here's the thing, ultimately all of the neural net stuff is just about analyzing statistics from sets of images, and then manipulating, or perhaps generating new unique images, based off of the aggregate statistical properties of the database of images used for training the neural net.
Neural nets are universal function approximators, so that is why they end up being great black boxes to train on real world data (images in the case of the things visual artists are interested in). they can adaptively learn to model the data by just being exposed to , and trained, on that data. Data again being collections of images when you are talking about what everyone who reads this users forum is interested in.
Neural nets are designed to loosely model what is going on with collections of neurons in the brain. Hence the term neural.
Ultimately it's just another tool for the artist. One that is a hot fast moving research topic. So people get excited, the press gets excited, one can develop a certain amount of 'neural net fetish' if your work is associated with it. But at the end of the day we're all trying to create visually compelling images. How we got there is less important than where there is, or maybe what there looks like.
Your last paragraph says it all. Computers have allowed us to analyze data in a way previously not imagined. However to my way of thinking there has to be a way to transcend the data for creativity to manifest itself. The deeper one goes into and becomes conscious of awareness within one owns brain and self, the greater the creative response will be. Intuition is not just a silly word. Why be a prisoner to data output? Try the road less traveled.
You could use something like the Vectorizer to transform a natural image of people into something more stylized like the source used for the processing in the image above. And then load that Vectorizer processed stylized image of the people as the new source, and then run the vector paint presets on it to get the painted look for it.
Here's an example from my art blog of what i mean by vectorizer processing a natural source image into something more stylized. The source image is a totally normal photo of 4 people.
Don't ask me for the Vectorizer preset, because it was generated on the fly by gallery show in Studio Artist V5.5 during automated testing, so it's totally a generative art thing designed by the AI generative system.