AI Paintings - What Are They And How Can You Get One?

AI Paintings - What Are They And How Can You Get One?

 

AI paintings have been causing a stir in both the computer intelligence and art communities in recent years. But what exactly are they, how are they created, and are they really art? Keep reading to find out.

What are paintings generated by AI?

AI generated paintings are exactly what they sound like - artworks created by artificial intelligence. That means an algorithm programmed into a computer generated them. There are two main ways that looks.

One way AI learns how to do things such as generate art through a process called machine learning. This means that the system is given a set of data - in this case, artworks - and it learns to understand the similarities of these data before setting out to reproduce what it knows about artwork.

So, if the AI is fed a series of images from the renaissance period of art history, it'll eventually learn and recreate a piece of artwork that looks like a renaissance painting.

Many people have taken issue with this - citing that it isn't really creating art, but simply regurgitating it.

The other method is one which doesn't focus on guaranteeing cohesiveness with a genre, but one in which the artificial intelligence realizes what makes up the core components of the art it's processing, then uses what it knows to create a piece similar but with a distinct style. 

The latter is the technology used by the engineering team on super-computers at AI Art House.

Are artificial intelligence paintings really art?

Well, that's for you to decide. Most people's opinion on this depends on the type of algorithm used to generate the art.

Earlier we talked about machine learning, and how there are two main camps when it comes to this approach to creating AI artists. Well, each of these things has a name.

When AI is given a set of images to process and recreate, this is known as a Generative Adversarial Network. It uses two neural network forms (processes of thinking similar to a human brain); one generates new things, the other checks the product against what it knows to guarantee similarity.

The alternative process, where an AI develops its own technique, is known as a Creative Adversarial Network. Instead of checking for similarity, the neural network aims to create a piece that differs from the genres it knows.

When 'Edmond de Belamy' sold, many people in both the art and artifical intelligence communities were upset. Many believed that simply learning to recreate images does not prove computer creativity in any way, it simply demonstrates an ability to understand the visual components of an image and methodically program them into a new picture.

Some think the CAN is more promising as signs of true creativity from AI, as it begins to build its own style. However, it's still dependent on an initial data set of pre-existing artwork, rather than creating its own from scratch.

To some, the fact that AI generated art currently depends on machine learning and is not spontaneous/entirely autonomous means that it can't be deemed as art, as it's still relying on what its fed - which is subject to the biases of the human feeding it data.

In this sense, people think that AI needs to be truly autonomous, independent of a data set, in order to be considered truly creative and capable of making 'art'. AI currently can't do this, but perhaps in might in the future.

What was the first piece of art made by AI

Edmond DeBelamy

It's hard to trace exactly the first piece of art ever made by AI, but the first to ever go to auction was 'Edmond de Belamy', which sold for $432,500 at Christie's in 2018. It was created by an art collaborative in Paris, formed by Hugo Caselles Dupré, Pierre Fautrel, and Gauthier Vernier. The three are better known under the name Obvious.

Prior to Obvious, there was AARON, a program developed by Harold Cohen at UC San Diego in the 1970's. It is an autonomous system which creates bright, colorful art - similar to that of Cohen's.

A Google report, released in 2012 by Andrew Ng and Jeff Dean, announced that they had exposed a neural network to millions of screenshots from YouTube videos. They wanted to see whether the neural network could learn to tell the difference between images and start grouping them together. One took a particular liking to cats - it responded well to them, and learned to group them together.

In 2013, Simon Colton created the Painting Fool. This program learned by reading articles about the Afghanistan war in the Guardian and looking at the images used in each article. Using what it knew about emotion extracted from these articles, as well as the images used in each one, it created an image representing the war.

Colton said the reason for having the program learn and utilize emotion as well as images was to prove that computers would have to do more than simply pass the Turing Test to be labelled as being capable of 'creativity'.

The Google Deep Dream program gained attention in 2015 when it produced artworks based off images of skies and abstract shapes. It used the 'inception' algorithm, which zoomed in on images to understand the relations between pixels that create an entire image.

Google Deep Dream uses what's known as a Neural Style Transfer rather than a GAN or CAN. This means it recognizes a style of art, and when given another image, turns out a new image that recreates the image to fit the style of art it knows.

It works similar to filters on Snapchat or Instagram, which makes some people skeptical as to whether it can really be given the label of AI art.

2019 saw the establishment of AIartists.org, a website that brings together artists using AI in their work and chronicling new developments in the field.

Hybrid Paintings Created by AI

Lately, many artists have taken to using AI to assist their painting and other artwork. This means that the AI isn't creating the art entirely on its own - it's being used in collaboration with the artist, who may take inspiration from the AI art or use its technique as an embellishment to their own style.

AIartists.org is a place that centralizes AI artists - both hybrid and those solely using AI - so you can discover your new favorite artists and styles.

The other thing to remember is that there is more to AI art than AI painting. There's also AI music makers, dancers, and other visual performers using this awesome tool.

AI Painting Example

How to Make AI Art

Fancy making your own AI art? It might not be as complicated as you think if you were to go with the simpler method of reproduction.

If you know how to code AI algorithms, then go ahead and create your own based on whatever art style and machine learning best suits you.

If you're here, though, chances are you don't have this type of skill. Don't worry - these programs will have you up and running in no time to make your own visual art with artificial intelligence.

Artbreeder

Art breeder is the ultimate easy exploration in AI generated art. You can start by uploading two photos; the machine will mash them together. You can select the most interesting photo it generates, and each time you select, more possibilities will be opened up.

The mixing of two images together means you can create portraits, including anime portraits, as well as landscapes that look like they're out of a 4k game, as well as the type of abstract art we've been talking about.

You can select what style you'd like, as well as altering the 'genes' of each photo until you get a final image you're satisfied with.

ML5.JS

This program gives you an API with full instruction on how to use it and how to get started. It's helpful if you know a little code beforehand, but not essential.

It's not a program focused entirely on AI art, but more on the general process of machine learning - and on how you can learn to program it.

However, it does have a CAN feature, which allows you to create new artwork from subsets of photos you feed to it.

It also has a feature based on Google Quick Draw which creates little doodles using a neural network.

Google Deep Dream

We've mentioned this one a couple times before, but DeepDream is free and easy to use. It's a great way to get started if you don't know anything about AI software or coding.

It has three features, in which you can transform whatever photo you upload. These are deep style, thin style, and deep dream.

Deep style uses your photo and a given style of art to transform your image using the techniques it's learned from the pre-existing artwork. Thin style is similar but won't have quite the same transformative results.

Deep dream is when the machine alters the image based on what it has learned from zooming in on the image. It creates strange, psychedelic-looking recreations.

Will paintings made by AI take over the art world?

Who knows! None have yet sold for a price quite as high as the infamous Edward de Belamy, which to some indicates a decline in the interest for AI fine art.

But, at the same time, computer generated and AI hybrid art are definitely taking off. The amount of artists and art works now being made either fully or in part by artificial intelligence is at least in the thousands.

It seems likely that, as AI develops and becomes more prevalent, so too will AI generated art. This is simply a logical step in progression.

However, it also seems unlikely that artificial intelligence will completely take over the auction house. Human art is highly valued, and likely always will be - unless, of course, AI begin to outsmart us and don't take favorably to art works generated by human beings.

AI and human creativity produce very different types of visual arts from one another, and for different reasons. Because of this, it seems likely that the two will eventually run side-by-side, rather than in competition with one another.

Final Thoughts

AI art is generally relatively new, with AI artist photos and painting selling for at least a few hundred dollars. So, if you're thinking about getting in on the AI art bandwagon, now is a good time to join.

If you or someone you know has the coding ability to write a GAN or other algorithm to harness machine art, it's worth pursuing.

Not only is its price high, but it's also fun! Getting to know how to use machine learning, its limits and strengths, is fairly rewarding - especially if you do it from scratch. Even if you use one of the codeless programs we recommended, AI art can still be a great way to pass the time - something we all need right now!

AI art is developing rapidly and we can't expect to see it disappearing any time soon. However, there's also no need to worry - we will probably still admire human painting for many years to come!