Non-Fungible Tokens (NFTs) are all the rage at the moment. I will spare you my thoughts on this specific topic, but it did inspire me to get into generative art so not too bad of a thing really.
First of all, what is all this "generative art" malarkey. Well, basically it is the process of enabling a computer to generate some, or all of an image. This is not the same as using a computer to help you draw something - like Photoshop, as 'you' - the 'artist' - are very much in control of nearly every aspect. Instead, the computer gets to run the entire show. This does lead into some very dark places around copyright and "can a computer be creative" but we will ignore these points for the purposes of sanity.
I have focussed on 3 main areas of generative art so far.
- Functional modelling - using maths and physics to generate images of things like particle fields (above left), or nature inspired geometry;
- Full generation of images from noise - an Artificial Neural Network (ANN) system will start with 'noise' and generate an image it thinks you asked for - the space ship above;
- Stylising an existing image - an Artificial Neural Network system will start with an image you supply, and then augment it using the styles it learned elsewhere - me in the style of Piet Mondrian above right.
The final image can only be part of the result. Saving each frame as it's generated and you can stick them together into a movie. Here is a 6 minute long showcase of images being transformed and generated by Artificially Intelligent Neural Networks. Grab your popcorn and check this out:
So now you're suitably inspired and want to invest a few million in my efforts... thanks, but hold your horses. This is coming soon. In the meantime, how's about some simplified technical stuff about what's going on?
Functional Modelling
Maths and physics can be tricky, but nature uses it all the time and it can be truly beautiful. Water wreathing patterns in rock, seeds in a sunflower and Romanesco broccoli are all things you can see easily. But things that are hidden can be wonderful if made visible. You probably remember playing with a magnet and iron filings as a kid. Modelling these things and adding some artistic flair can produce some stunning imagery. The best part is that the process can be made into an actual process and each frame reveals a little more. This does lead to the question of ‘if generative art is digital, so should the output remain digital?’ The great thing here is that it really doesn't matter and it is up to each of us to decide how we can enjoy this artform. Printed on canvas, desktop background or dedicated display hardware are just some of the ways to interact with generative art.
Artificial Neural Networks
Neural Networks are modelled on how our brain works. Neurons receive an input, fire in a trained way and pass that on. Grouped together into networks they can be trained to perform many tasks from recognising and classifying things to modelling complex systems. Training occurs in the same way humans learn - by doing - they try something, take on board some feedback and try again. For the purposes of this discussion we will focus on 2 types.
Generative Adversarial Networks (GANs)
Two Neural Networks are used in this approach, the first is already trained on how to recognise and classify images. Generally, this has been trained with image sets that are widely available with lots of meta-data, for example, Wikipedia and Google image search. The second network is then 'trained' to generate what you have asked for and is marked by the first network. This is done by starting with the last frame (or complete noise) and then tweaking the image a it to see if it's getting closer. Because of the iterative nature of this process, you can start with any image and let it run to stylise something or generating something unique from scratch. The image of me was stylised in this way. The defining features of all the artwork by Piet Mondrian were know and my image was iteratively tweaked so that it demonstrated these features. Due to the way the human brain is actually wired to recognise faces, interfering with images of faces can go either way.
Diffusion Modelling
Not that other network techniques are easy by any means, but diffusion modelling is very processor intense. If you have ever played with the blur function in Photoshop, or tried to sit through a recent apple launch event, then you are familiar with "diffusion" - turning an image into a blurry mass of pixels. A neural network is trained to do this in reverse. The input image is broken down into feature areas and worked on individually, then the whole image is tweaked to blend it together. While the upside of this approach is that you can work with bigger pictures, the downside is that it takes significantly longer to generate. Even on the top end power platform from Google a single image takes about 10-15 minutes to generate.
What next
New models are popping up every day, some generate photo-realistic characters for game engines, some combine methods into some very interesting video effects, but I think that as far as art goes, the fad will pass and people will enjoy art as they find it. Generative art will become just one of the accepted ways to enjoy images. Nobody truly understands what Jackson Pollock was meaning, but that's ok, some people enjoy the lines and colour. Forget about the pretentious pompory - if you enjoy an image, enjoy it and don't let anyone get in your way.
Interactive generative art installations are now becoming a thing as well. An art piece can be influenced by using cameras proximity sensors or heat sensors. These can take the form of music performed while you wave, flower blossoms falling as you swish, or taking an image and transforming it based on where you're standing and the weather - all truly unique pieces that cannot be replicated. Maybe you will have an interactive display that will show you an interpretation of your day based on your calendar, or try to lift your mood by showing you a different you every morning.
The future is uncertain... but it's going to be colourful.
No comments:
Post a Comment