From Harry Potter to Breaking Bad and Lord Of The Rings: who needs a Kim Kardashian when there’s new fictional muses to dress in Balenciaga. Generative AI is a hot topic and is currently taking its integration course for our society. In this article, I’ll be diving more into AI generators, algorithms and prompting – flavoured with my own unsolicited opinion and the latest gossip in tech.
Davy de Lepper | 25/05/2023
Last time I took you through the factual fairytale of generative AI and its entry into pop culture. Through viral examples – such as the “Drake & The Weeknd” case – I discussed the opportunities and dangers of new AI techniques that seem to have suddenly appeared out of the blue. We are now three weeks on and generative AI is still the topic of conversation on the tip of everyone’s tongue. That’s why, in this article, I’m examining the production process of AI art using illustrative cases from my social algorithms with the help of artificial friend ChatGPT. To what extent can we still speak of originality when an image is “painted” using existing material on the web? But first things first: what is prompting? ChatGPT, take it away:
In the context of AI, “prompting” refers to providing a language model with a starter phrase, instruction, or sample text to direct text generation. It serves as a trigger for the model to produce the desired output. Prompting is a valuable tool for controlling the output of language models and shaping their generative capabilities.
The evolution of media
One of the most controversial AI cases in recent times was that of Boris Eldagsen. The 52-year-old German artist won the prestigious Sony World Photography Award (SWPA) with an image titled The Electrician. The AI-prompted photo in the style of Roger Ballen‘s psychological portraits – a photographer who became known to the masses through South African pop duo DIE ANTWOORD – won in the Creative Open Category. Sony allowed Elsdagen the prize, but denied him the debate he aimed for. Even so, his deliberate move made almost every headline and opened the conversation about AI in the classic perspective of what photography is or should be.
The Eldagsen case is now old news, but still relevant as an example illustrating how classic media changes over time: the current phase of change being marked by AI playing the leading role. The image that Eldagsen submitted for the SWPA was created using DALL-E 2: an image generator by parent company OpenAI, which marketed ChatGPT, among other things. The answer to the overarching question – is it still photography when it’s AI-generated – remains personal. Eldagsen himself calls it promptography. As a classically trained documentary photographer, where one has to hunt for a subject in front of the lens in the real world, I agree with Eldagsen name for a new kind of photography. It makes me curious about how I can generate such an image myself. To get my hands on this knowledge I enlisted help from my new friend, the Google-with-an-opinion and AI entity ChatGPT. I asked them the following: how is an image created by an AI generator after a person enters what he wants to generate?
There are several methods that AI generators can use to produce images, including the use of neural networks and generative adversarial networks (GANs). Neural networks operate through layered processes where information is passed through a series of neurons, while GANs consist of two competing neural networks that work together to produce images similar to the training data. In short, AI generators produce images through complex algorithms that learn patterns and characteristics from training data, then apply them to new data to generate an image similar to what humans input.
The answer doesn’t generate a “eureka” moment in my tiny human brain, so I let my Instagram algorithm seduce me and delve into the generator of generators: Midjourney.
A journey through Midjourney
I started on PromtJungle’s YouTube channel, which instructs in a tutorial how viral videos like Demonflyingfox’s Harry Potter x Balenciaga video are made. My first revelation comes when ChatGPT and Midjourney are combined for the creation of these videos. My brain continues: could I dress stiff Dutch Prime Minister Mark Rutte in the avant-garde punk of recently deceased English fashion legend Vivienne Westwood? I gave my ironic hunch a try in the step-by-step tutorial that spawned several viral videos and ended up with the following prompt:
1990 screengrab of Mark Rutte wearing a revolutionary Tartan Suit paired with a black leather belt adorned with silver spikes and chunky platform boots, fashion scene Vivienne Westwood commercial –ar 3:2 –v 5
The next step was to figure out platform Discord and get my hands on a Midjourney subscription. After finding the right channel and putting in the prompt above, the results that generated in front of my eyes barely resembled the prime minister. After adjusting the prompt and adding a portrait, sponsored by Google, Midjourney generated 4 different flavours of Rutte that didn’t even resemble one another. This proves that even Midjourney is not yet completely consistent in its database, because only the most famous figures of our world will be hyper realistically imitated. Of course, this will change in the coming years, maybe even months, when you look at what the AI of app Lensa is already capable of with some selfies from the gallery on your phone. Below one out of many results that I prompted:
Promptography, or just prompting, is a trend to stay. However, one may wonder whether the AI technique with which ‘digital art’ can be generated is inferior to the creativity of makers. My Midjourney-prompt-experience may have been a little less smooth because I didn’t pair a blockbuster movie with Balenciaga, but it did show me just how narrow a prompt’s language should be and what else it can be used for. For example, you yourself coud try generating a digital self-portrait, 3D logo or interior in your own signature: prompting is not just for viral YouTube videos. Like ChatGPT, make it your own and explore the off-the-beaten-track capabilities of the programme. My Mark Rutte meets Vivienne Westwood look may not have been completely successful, but it makes me curious how I can apply Midjourney as a designer for fashion and styling assignments.
Prompting is just a small taste of what’s possible with AI in art. So in the next article I’ll be taking a closer look into AI art, highlighting artworks by key artists, here on Nxt Museum’s The Flash Drive.