What is generative art?
written by revdancatt
This is the (lightly edited) transcript of the video I made over on YouTube, embedded below, for those that like to read at their own pace!
Often we think of art and artists like this....
An artist creates some art, be that with paint, pens, pencils, clay, or various other materials.
However with generative art, and artist creates some instructions, which often contain a variable elements, someone or something will follow those instructions to create the art.
And that, is generative art.
Hi, my name is Daniel Catt, I'm a generative artist who uses code to create both digital art, and physical art via a drawing machine. This is my take on what generative art is near the end of 2022. I'm going to break this down into a number of sections, which you should find down below, so feel free to skip around if you want. As an experiment I've also created this text version of the video with the images I used if you prefer to read at your own pace, let me know if you like this format too.
Anyway, We'll start with a look back at a few examples of studio assistants in art, which I think will really help place generative art into the context of art. So...
STUDIO ASSISTANTS
We're going to look at instructions and people carrying out those instructions. If we go way back in time to the "old masters" you would have apprentices or pupils who would do quite a lot of the groundwork on a painting for the "master" to then come in and finish off. Rembrandt for example was a student for several years, before becoming a teacher himself.
To start with, an apprentice could be set the tasks of painting in the backgrounds of a scene, they were trusted to do the sky and some hills. But as the apprentice improved they were asked to do more and more of the foreground work. At the end of an apprenticeship, the student's work was often indistinguishable from that of the master, which has obviously lead to some interesting situations.
So we have this idea of an old master painter painstakingly producing one work after another, while the reality is that it was often more of a factory, where eighty or ninety percent of a painting was painted (under guidance) by studio assistants, with the final touches being added by the artist.
You would have Artist, Instructions, Assistances, Artwork.
Bridget Riley, would conceive and design an artwork, then have one assistant draw out the lines, and another then paint black and white (and later colour) within those lines. Bridget would then sign the artwork down the side, and that became an original Bridget Riley artwork.
Like many artists today, Miss Riley employs assistants: “The actual execution takes so long that it would be frustrating to do it all myself. It’s self-evident that my art doesn’t rely on the actual handling of paint.”
Two more...
The Comedian, an artwork by Maurizio Cattelan, is a banana taped to the wall with some with some duct tape. It has the very specific instruction that the banana must be taped 175cm above the ground and that it should be replaced every seven to ten days.
It is not unusual that artwork, especially works that decompose over time, or involves lots of heavy materials to come with instructions so the gallery can recreate the artwork, rather than attempting to transport it, which is often impossible.
In the case of The Comedian, what's actually for sale is the certificate of authenticity (which I'm going to call the certificate of ownership now and then because I can't pronounce authenticity easily) so one of those and the set of instructions. We'll come back to this one later.
The last one I want to cover, which is closer to our generative art is Damien Hirst's Spots. Damien's studio assistant at the time, was the wonderful artist Rachel Howard. And Damien's instructions for creating the Spots was roughly, "use household gloss paint, never the exact same shade twice".
So here, the instructions are allowing an amount of variance into the final result. He's not specifying the exact colour, just that they need to be mixed so you don't have the same shade of yellow appearing more than once in the painting. Hirst has this to say of Rachel...
"The best person who ever painted spots for me was Rachel, she’s brilliant. The best spot painting you can have by me is one painted by her."
The interesting thing here is that Rachel could create a Spots painting in the studio and it would be given a certificate of authenticity. But the exact same person, Rachel, could go home, use the exact same materials, follow the same instructions and create arguably one of the best spot paintings, and it not be an original. Because it doesn't have a certificate of authenticity, that makes an artwork genuine or not.
We're going to touch very briefly on "Information Theory" now, it'll be fine.
The particular aspect I want to cover is the transmission of data. It is easier and faster to transmit the instructions to make some artwork, than it is to transmit the final artwork. The instructions to generate one thousand artworks is easier than shipping the one thousand artworks.
Say you have some cloth made in Italy in the early eighteen hundreds, and you need a lot of it in England. Well there's a time and cost associated to transporting all that cloth. However, instead if you have a Jacquard machine connected to a loom in Manchester, and you transport the punch-card instructions from Italy to England you can then create the cloth to the Italian designs as you need it.
Interestingly if you shuffled the punchcards, if you rearranged the instructions, then you'd get remixed versions of the patterns.
Transmitting a play, a performance, is a similar thing. You don't send all the performers from one country to another, you transport the script. Or a piece of orchestral music, the sheet music, the instructions are sent, and then the final piece is performed by a whole new orchestra.
That's the bandwidth and storage of information that expands out to something much bigger in a nutshell. Back to our modern generative art, and the instructions.
So, we've established that having an artist create instructions and then someone or something else executing those instructions is a well-established method of creating (and transmitting/transporting) art. We're going to look at three different important flavours/types of generative art.
First...
DETERMINISTIC
...which basically means, we as a human can understand and predict how the instructions will change based on whatever the inputs are.
Take our orchestra and sheet music for example. Say we do something fancy like having four-page sixes, labelled Spring, Summer, Autumn and Winter, and when the orchestra plays they swap out that page depending on the season. Easy! Let's add some more rules.
We split the world up into five latitude bands, around the equator everything stays the same, then above or below 20 degrees latitude a few pages are played one octave up or down. Above or below 70 degrees they're played two octaves up or down. And we'll split the world into eight segments around, and, I dunno, transpose some other part of the music based on which segment you're in.
Now, suddenly, the final composition performed in London is different than it is in New York.
We'll throw two more rules into the instructions, first we have another selection of pages that get their order rotated or swapped around based on the day of the week. And finally, on a Full Moon or New Moon, we do something else musically clever and interesting, I dunno.
These variations are both easy to understand and completely predictable. If the piece is performed on a full moon, on a Friday, in September, in New York, we can know what the sheet music will look like, we could even perform it, but we have to actually be there at that time to experience the "original/authentic" version. And to experience that specific performance again, you would need to be in New York, during Autumn, on a Friday with a Full Moon. There may be some years when that doesn't happen at all.
We know the overall form of the performed piece, but there are a number of variations, which are completely predictable, and deterministic.
Now onto...
RANDOM
The opposite of deterministic is unpredictable randomness, I'll give an example.
There's a very simple "generative art" program called "10Print" a quick google will find a bunch of examples. But it's basically this line of code.
10 PRINT CHR$(205.5+RND(1)); : GOTO 10
All that's doing is either printing a character that's a diagonal dash that goes one [\] way or another [/], I'll reproduce it here using a neatened-up diagonal line.
As you can see what we end up with is a fairly complex-looking pattern, it looks like there's some kind of design behind it as patterns emerge, but it's literally just a single tile that's placed either one way or another.
If we swap out the design for a couple of curves, we can see another complex-looking pattern which follows a very simple rule.
We can reproduce this by hand. If I take a grid and a coin or a dice and then manually draw in the lines one way for heads (or odd) and the other way for tails (or evens) then I can create a pattern by following those instructions.
Now, these all obviously belong to the same group, they look similar, even though each one is unique. If we destroy one, and if we didn't have this image of it, there would be no way to recreate that very specific output, because it was purely random. The flip of the coin, or the roll of the dice is unpredictable.
This brings me onto...
DETERMINISTIC RANDOM
Which sounds a bit odd, and is easier to demonstrate than explain, which I'll do in a moment. But essentially, it's a sequence of numbers that follow a rule to generate them, but you or I as humans couldn't predict the next number in the sequence, even when viewing the previous selection of numbers.
For example, if I give you this series of numbers: 8688003716073032205474 which are near the seven thousand digits of π there's no way for you or I to really predict the next number in that sequence is two, even though it definitely is and always will be. Next time you see the sequence 8688003716073032205474 then you know the following number is 2, but now we're stuck on the next one after that. It's unpredictable by you or I but completely deterministic and repeatable.
Together, we can now make same generative art, based on 10 print, with these instructions.
- Think of a number between one and ninety nine excluding ten, I'll pick twenty-three.
- Take an iPhone, open the calculator app and flip it sideways.
- Hit the π button.
- Multiply it by the number you thought of.
- Take the first thirteen digits after the decimal point, and loop over them.
- For an odd number draw a diagonal one way, for even draw it the other way.
Now we know three things.
ONE: If I throw that artwork away, or destroy it, or whatever, as long as I remember the instructions and the number I selected (twenty three) I can recreate that artwork at any time, and it'll be the same.
TWO: If we both pick different starting numbers, our results will be different.
THREE: Unless you're a super maths genius, if I tell you the selected number is forty two, without having already seen the results it's unlikely you can predict the final artwork, meaning it's as good as random, for us humans it's unpredictable, while being completely deterministic.
Generative art is basically that, but with more rules and more complicated 😁
So why the explosion in generative art right now?
I'll cover this in more detail in a later video, but the "old way" to do generative art was this.
An artist would write some generative art code, that created thousands of variations on a theme and it could be completely random as there was no need to be able to reproduce an artwork.
The artist would keep creating outputs until they saw one they really liked and they would carry on until they had a selection of what they considered the very best results. They'd then narrow those down to three, maybe five, and...
Make prints of the results!
Someone would buy a print, which would then be stuffed into a tube and flown halfway around the world. Which isn't, you know, great.
But then recently NFTs came along as a "solution" to selling digital art, without all that printing, tubes and shipping things all over the globe. Now making it feasible to sell the outputs of generative art. Instead of an artist picking three or four to sell, they can create a set of instructions that can create fifty, a hundred, a thousand results, which people can buy.
Now, I'm not going to get into NFTs, but it is important to put modern generative art into the context of NFTs, so I'm going to give a very generalise overview.
The first thing to know is that the NFT is NOT the artwork
The NFT is the certificate of authenticity, the certificate of ownership. To make it easier to understand if we go back to our banana taped to the wall, "The Comedian", as an example.
Say for the moment, collector Anna has the certificate of ownership, and is showing The Comedian in gallery A, gallery A is using the instructions to present the work. Meanwhile, gallery B in a different country is also using a copy of the instructions to present the work.
The banana in gallery A is the "real" one, and the one in gallery B is the "fake" one. Someone (Bob) walks into gallery B and is so taken with the artwork they want to own it, they get hold of Anna and buy the artwork off them for a huge amount of money. What they're really buying is the certificate of ownership.
The certificate and original set of instructions are flown directly to Bob. who now gives gallery B permission to show the work.
The banana taped to the wall in gallery B is now the "real" artwork and the one in gallery A becomes the "fake", even though they haven't physically changed. The only thing that's changed is who owns the certificate of authenticity.
So in the case of an "NFT artwork", people can "Right Click Save" all they like, and that's the great thing about it. With a physical artwork someone could buy it and then tuck it away to "keep it save", and then no-one has access to it. Copies of an "NFT" can exist all over the place, everyone can enjoy it. But, if you care about such things, and many people don't, ONE person "owns" the artwork because they are the only person with the certificate of authenticity. The NFT is the certificate, that says who owns the artwork, it's not the artwork itself.
Okay, so back to generative art. Let's have a look at Fidenza for an example. This is a generative art project by an artist called Tyler Hobbs, and it came in an edition size of one thousand.
Before the project launched the artist showed a selection of example outputs, so people can see the overall style of what their set of instructions (the algorithm) can produce, but no-one knows what EXACTLY what they'll look like until they're created.
Someone buying a generative artwork, from a "long form" generative art project (and I'll cover that in another video), is buying into the overall aesthetic of what will ultimately become the collection, but they don't know specifically what they're going to get.
Imagine our 10 PRINT example from before. Instead of thinking of a random number between 1 and 100 you say, "I would like to buy one of those artworks" a unique number is created for you, put through the instructions, and then the art is generated.
Your NFT is the recipe (the proof of purchase), the certificate of ownership AND also contains your unique number.
At any time, you can feed that unique number into the instructions and re-generate the artwork.
If we take Fidenza zero for example, it's unique number is... on-screen, because I am not saying that! [0x55ac31f9309a8a914ab889fca907f350321024a60d39670cbb693c9c14638d94], Every time that number is fed into the code, it creates Fidenza #0.
That Fidenza can be displayed all over the place, but only the person who owns the certificate, the NFT, is agreed to be the owner.
We are NOT going to go down the rabbit hole of agreed ownership in a society.
For fun, if we change the very last digit from a four to a three, then we get a whole new Fidenza, but no certificate, no NFT exists for this new one, you can't "own" this one. There are only one thousand OFFICIAL NFTs for Fidenza, anything beyond that is a fake.
One last thing and then we'll wrap up with the summary.
This comes up sometimes so I figured it was worth covering.
The "blockchain" is like a giant ledger in the sky, it's a distributed database that you can write information to once, but you can't update it, you can't change what's already stored there. As an example, I want to commit to the blockchain the fact that I like cheese, so I ask a number of people who are part of this distributed system to write the data "Dan Catt Likes Cheese" to the blockchain, and I have to pay them a small amount for that.
The fact that I like cheese is now immutably on the blockchain, it's publicly visible and anyone can read it. I now can't change that information, it can't be updated. What I can do is pay a second time to write some more information, "Dan Catt Doesn't Like Cheese". Now everyone can see that I paid for both sets of information be written to the blockchain, in which entry is the most recent.
One thing to remember is that it costs money to put things onto the blockchain, the bigger the thing the more it costs. Something like "Dan Catt Likes Cheese" is very little data, but a whole high-resolution image would cost a small fortune to put onto the blockchain.
We've heard those stories about someone buying an NFT (which is just the certificate) that points to an image that's stored somewhere "off-chain", and then that image disappears. So the owner is still the owner, but the NFT no longer points to anything.
So going all the way back to our Information Theory, the transmission and storage of data. The instructions for creating the art, is smaller and takes up less space than the final image, let along a thousand images.
By placing the Fidenza code, and my own 70s and 80s Pop code, onto the blockchain itself, the same place the NFTs live, then it means that it's always verifiable, people can check that it was actually me putting that code onto the chain, they can even verify when I put the code onto the blockchain, and also how much it cost me, in this case one thousand eight hundred dollars for my first 70s Pop project.
When someone pressed the buy button, a unique number is created (the hash), it all gets wrapped up in an NFT placed on the blockchain, the hash is fed into the code, which is also on the blockchain and the image is created.
If Bob comes along, and we’ll pretend that Bob is standing in the distance because of Covid, and buys the artwork from the current owner, a new record is added to the blockchain stating that Bob is now the new owner, and everyone can check and verify that’s what’s happened.
It doesn't matter if that image gets deleted or lost, because the code exists on the same blockchain as the NFT token and you can always feed one into the other to regenerate the art.
ArtBlocks, where I release some of my projects is 100% On-Chain.
Meanwhile fxhash, where I have some other projects is Off-Chain. The NFT, the certificate of ownership, lives on the blockchain, but the code lives off the blockchain, so should the code ever vanish (from everywhere) then you won't be able to recreate the image again should that also vanish and you lose it.
The advantage of having the code off-chain is that it allows artists to have a lot more code, and also include images, audio, even video, because they're not restricted by the cost of putting things on-chain, the more data the more it costs.
So this is why you can't really compare platforms like ArtBlocks and fxhash because they're doing things in two different ways, on one you have to be a lot more conservative with the size of the code, the other you don't have that same restriction, resulting in different types of art.
Anyway, in summary. What is Generative Art?
Generative art is when an artist, often using code but not always, creates a set of instructions that takes in a unique input, and uses that to make a whole bunch of decisions to generate an image.
Feed in the same input and you'll get the same output. Feed in a different input and you'll get a different output.
Hopefully that's a good enough introduction to where we are with generative art, and why it kicked off and then settled down a little bit near the end of twenty twenty two. I'll probably do an update to this in mid twenty twenty three when it's all changed again.
Subscribe if you like this sort of thing, hit the like button if you found it useful, share it with people when they ask "Hey, what's generative art?", show it to your students in a class so you get half an hour off, and get them to do their own version of the 10 PRINT instructions OR come up with their own rules. Drop a comment if you disagree with anything, we love that! And I will see you in the next video.