Rival Consoles - Overflow
June 28, 2022
Art offers humanity so many vital freedoms. It allows us to document our surroundings and our lives. It enables us to celebrate the communities
that we are a part of. It encourages us to make bold statements, invoking emotions that we feel will best convey our point or idea. It also offers escape. A chance to create worlds that differ from our own in pursuit of truth, beauty or merely entertainment.
Separating a piece of art from its creator is no new phenomenon; art can be enjoyed on its own merit. But in recent times, a movement has occurred that distances the artist from their creation in a very unique way. The space carved between the two is filled with thousands of photographers capturing millions of images, neural networks that battle against each other and reams of dizzying code. Though it may sound in opposition to the traditional creative process when expressed like this, the movement is responsible for producing incredible artwork and indescribable visions. This is the spectacle of images generated by artificial intelligence.
AI is a part of our world and it is a part that continues to grow in size and stature. Despite this, its processes are incredibly intricate and often too convoluted to allow for a clear understanding. With this being the case, it should be no surprise that the systematics of AI artwork are complex as well, with a plethora of different, intricate methods.
The majority of AI artwork created using algorithms known as GANs (generative adversarial networks). This is a type of framework in which two separate neural entities – one generative and one discriminative – vie against each other. Given a dataset as inspiration (in the case of art, this would usually be a set of photos or images) and a worded prompt, the generative entity begins to layer pixels. As it builds an image, the generative entity must trick the discriminative entity into believing that the iterations of the word prompt it is offering have not been digitally synthesised and are, in fact, the real thing. Its task is to create something as identical to its composite understanding of the prompt as possible. The generative entity rapidly produces image after image for inspection, chipping away ever so slowly at the blank marble of pixels until an approximation of the prompt slowly begins to appear.
Approximation is a keyword in AI artwork. There are many processes in which neural networks are used to refine an image in aid of human endeavour. For example, processes such as inpainting (where damaged or deteriorating photos are retouched) and large scale agricultural operations in which AI technology is used to approximate levels of flooding in crop fields or what may be under cloud cover in highly dense satellite imagery. AI artwork operates in a different way. In those two examples the system is given an explicit starting point, a piece of the puzzle at the very least, and it uses its system to work out the rest. In AI artwork the system works towards what is essentially an unreachable goal. There is no definitive visual representation of the word ‘dog’, ‘sunny day’ or ‘boat ride in Venice’. But the machine must try.
A handy characterisation of what exactly AI artwork is, could be: An uninterrupted artistic process that can only be performed by a machine. Though the artist can tamper with controls and choose what iterations of the prompt they want, they are not able to manipulate the GAN’s process; the magic happens in a place that can’t be reached.
The Analytical Engine is widely renowned as the earliest version of what we now know as the computer. English mathematician Charles Babbage designed it in 1837, but his aspirations never took physical form due to financial reasons. Though Babbage’s revolutionary design has more to do with the general computer than any automated artistic process, it would be Ada Lovelace’s notes on the Analytical Engine that would explore possibilities far beyond simple computation.
She posited that there could come a time when computers would be used to create art and music. In an age where rationalism and creativity were seen as being the furthest things from each other, she spoke of the two coalescing in a way that the engine “might compose elaborate and scientific pieces of music of any degree of complexity or extent”.
It is difficult not to think of those neural networks birthing fantastic images when reading Ada Lovelace’s notes, especially when she mentions the Analytical Engine’s lack of true originality, saying “its province is to assist us in making available what we are already acquainted with”. The same is true of the GAN, though what we see may be awe-inspiring and even indescribable. It all comes from a dataset and a prompt.
Without a prompt, the system is lifeless. No revolutionary ground broken, no image materialised. But much more than a binary on/off switch, the prompt allows the user to create. To emboss themselves or their imagination onto the system before it gets to work. The prompt is the main creative entry point into the world of AI imagery.
There is a certain ease to a sentence-based prompt. For one, it allows the user to take a more overarching, conceptual approach to the creative process. An idea or vague concept for an image or landscape can take an exceeding amount of time to refine and bring to life, whereas the GAN system can produce an interesting iteration within a few hours. The prompt also means that the artist still holds a certain power. In their sentence, they can choose to invoke beauty or assert a lack of it. They can call upon a certain style of painting or time period. They can choose between intricacy and unkempt digital viscera.
This is beautifully displayed in an example provided by Vito Genovese, creator of the AI art project Genovese Generatives and owner of independent music label/project DMTFL. The artist offered the phrase “synthwave angel” to the system, and received many humanoid-looking forms and shapes, with wings also being a prevalent feature in many results. Aspects that were not as easily predictable were feathers of black and white piano keys adorning the wing shapes, and whispers of oceanic activity that lapped at the feet of many of the figures. This is the system trying to illustrate the words “synth” and “wave”. As Vito puts it: “You get to interpret what the image outputs look like, but the machine is interpreting what the text inputs should look like as well.”
The prompt does bring challenges, however. One being that it is quite hard to distill an imaginative idea down to one simple sentence. The lack of control over the shaping process also means that more visionary users may find their prompts interpreted in a way that they did not mean or plan for them to. Though this may not
be an issue for the recreational user, the more experienced among them may find it problematic when trying to create something with a specific look or feel.
The prompt has led to the artform being very inclusive, with anyone being able to flex their imagination muscles and see what comes out. Twitter accounts like @images_ai allow users to offer prompts and post up the results. A practice that has enabled followers to see many amazing images, from the artwork for Everywhere at the End of Time reimagined as a mystical fantasy landscape to collaborations between Van Gogh and H. R. Giger.
The prompt is also part of the reason why there is a lack of writing exploring the rich aesthetic value of the artform. This is not to say that there is not a community being built around the practice. As Vito says there is “a vein of excitement that has only just been barely tapped”. But when it comes to analytical writing, the belief seems to be that, due to the overarching computational nature of the process, there is no rewarding analysis to derive. But maybe there is a reason deeper than this, one that is tied to AI artwork’s fluid and vague appearance.
The most striking thing when looking at most AI artwork is its abstractness. It lacks the guiding hand of a human artist to steer it somewhere overtly purposeful. When studying these artworks for a very long time, one will start to grasp the inability of the process to create anything that is real. The image may seem like an accurate representation of the prompt at a glance, but focusing on specific areas of the piece will reveal that what is actually presented is a collection of shades and shapes that approximate the stimulus. There are no real analytical entryways into each piece, no parts through which to begin a journey toward an interpretive conclusion.
This is because the artwork lacks that form imbued by a human hand and mind that we are so comfortable with. An example of this is the somewhat grotesque AI-generated image that went viral in 2019. Lacking all the aesthetically powerful aspects of AI artwork, the image was not attractive, and almost frustratingly ugly. The subsequent discourse that occurred online was similar to other internet debates over misinterpreted images or misheard sounds. Only this time, there was no definitive answer. People expressed frustration, even disgust at the picture that seemed to be so many things, but was actually nothing at all.
This example illustrates how AI imagery unveils the fugue of the human mind, laying bare the frustration that ensues when we are faced with a process that is explicitly inhuman. It shows clearly that line between recognisable everyday objects and meaningless visual viscera.
The image may seem like an accurate representation of the prompt at a glance, but focusing on specific areas of the piece will reveal that what is actually presented is a collection of shades and shapes that approximate the stimulus.
This look of failure could quash fears of AI-generated art disenfranchising and harming human artists – a topic of debate that has come up frequently alongside naysayers claiming that the AI-generated pictures should not be classified as ‘art’. The golden rule to remember when exploring this debate, is that even the most sophisticated programs used to create AI are not autonomous; they do not work alone.
The lines between the two have been drawn fairly explicitly. When AI artwork is presented, and even sold, its automated nature is frequently marketed as its selling point. People marvel at these images and purchase it, not because it is any better or more impressive than human artwork, but because of the basis of its creation and existence.
The most well-known example of an AI-generated image being purchased came in 2018 with the sale of Edmond de Belamy. Smashing its pre-auction valuation of $10,000, the blurry, classical-inspired portrait ended up selling for $432,500. With the buyer being an undisclosed party, there is always that sneaking suspicion in the art world that the sale is not what it seems.
When taking this happening at face value, there is no doubt that the Edmond de Belamy piece sold on account of it being an AI-generated artwork. And though the image or ‘painting’ is interesting and beautiful on its own merit, the sale indicates more of a standalone landmark than some sort of cathartic moment for the artform or community. It was a bold and headline-grabbing spectacle that definitely put AI artwork in front of a larger audience, but not in a way that really benefitted smaller, independent creators.
What could be seen as more of a benefit to AI creators, however, is the NFT space. Vito explains that many artists have moved into ‘minting’ their AI works as NFTs, as seems to be in vogue at this time (or at least in late-2021, as interest does seem to be waning). NFTs are too much of a topic to fully explore here. But suffice it to say that though smaller artists have found success in the space, the lack of security and regulation, combined with the more prominent players in the market seemingly working more from a desire of scarcity to drive profit than artistic value to drive demand, makes it hard to believe that it represents a paradise for AI artists.
Due to its style, it is easy to see AI artwork as being an effective accompaniment to other forms of media, as well as more overarching explorations of artistic styles and touch points. That is where the practice neatly intersects with music.
A visit to DMT Tapes Bandcamp page finds vibrant AI artworks lined up like a miniature digital gallery. Each AI-generated square is a window into an interesting album or release. When listening to one of the growing number of projects, nostalgic sounds blare out over a uniform shroud of fuzzy feedback. While one listens to heady ‘80s power chords and gazes into a shroud of aquamarine colours formed into a bleary city skyline, it is hard not to feel as though the two artistic endeavours inform one another. Despite the lofi quality of the sample, the triumphant nature of the music is clear; despite the nebulous style of the AI artwork, that city skyline breathes and pulses with authenticity in time with the music. Vito explains that the music featured on the page in recent times is a transformation of the DMT Tapes name back into a solo project, featuring music and art by himself.
DMT Tapes was not always this way, having been a vaporwave label for around eight years it has been home to many different artists and styles, with varying album artwork. But as Vito made their forays into the world of AI, the connection seemed natural: “Vaporwave was a gigantic love letter to nostalgia.” The genre – or at least certain factions of it – thrives on the repurposing of music from days gone by into something different, and in many ways, something warped and intangible. Because of this, the two artforms can very easily embrace one another. Vito also points towards AI artwork’s ability to mimic certain styles, eras and even artists as a means to build upon certain musical movements and genres. The potential to apply it to projects like concept albums or releases exploring imaginary worlds are almost infinite.
Just recently, Genovese Generatives celebrated one year in the AI art space, compiling 300 GIFs as part of its Ultra Deco series that showcases images evoking the beauty and promise of pre-WW1 architecture. This is just one of many styles Vito has created through tweaking and formulating certain prompt patterns and styles. The images flow from abstract shape to shape; a unique sense of place and time washing over the viewer as new age-inspired music plays. The artist explains it best, as a “celebration of yesterday through the making of something new today”.
Technology and art are inseparable, both in terms of consumption and creation. But never before have the two embraced one another so closely, allowing for a kind of unbridled creation. AI artwork literally has the ability to show us anything that we can imagine, and its open source technology means that, by and large, anyone can plunge the depths of their mind. Genovese Generatives is a prime example of the possibilities available through the medium. The processes may be complex, but what it achieves was something merely pondered around 200 years ago and was even hard to comprehend until very recently. In AI artwork mind and machine entangle like never before, to form images of incomparable intrigue and bewildering rarity.