A Journey Into Generative Art: Part 2

March 23, 2023 Arts, Design, Technology, Science

Woman harvesting vegetables on a rooftop farm in a futuristic utopia. Daniel Simons. Created in Midjourney.

By Daniel Simons

In part one of Generative AI Will Change The World, we explored the disruptive impacts and ethical dilemmas surrounding the Generative AI revolution. In part 2 we ‘get our fingers’ dirty and take a deep dive into what it’s like to actually create Generative Art.

The moment I first became captivated by Generative Art was when I watched Jason Silva’s Shots of Awe video titled ‘My Thoughts on AI Art.’ The clip begins by reflecting on the uncanny moment in time we are living through, where machines are forcing us to reassess what it means to be an artist or art lover, then quickly becomes a rhapsodic celebration of the new era of ‘coupling between the biological and unbiological.'

As Jason ecstatically opines on how Generative AI is a magical tool that will let us ‘shrink the lag time between what we imagine and what we create,’ his words are translated into mind-melting AI-generated visuals in real-time.

This world-changing tech isn't the end of art, he suggests, as his face morphs into a hundred different incarnations, ‘It’s a ‘psychedelic fever-dream and a digitalisation of our inner life. It’s the exteriorisation of our innerverse.’ It’s hard to watch the video and not be awe-inspired, a little terrified…and utterly hooked.

As I gave more attention to Generative AI, I came to appreciate that I was in a ‘happening.’ The leaders of the tech world were suddenly comparing it to the advent of the internet, or the invention of the cell phone. It was set to be as inescapable as social media and just as replete with its own breed of unease and FOMO.

Just like with COVID, there was a vague, uneasy feeling that something fundamental had changed, but I couldn’t really grasp it. I needed to have a play.

My first foray into Generative Art was a complete disaster. I dabbled with a program called ‘Stable Diffusion’ via a website called ‘Hugging Face’ via a random link that popped up in my social feed.

I typed into the prompt box, ‘A woman in a park reading,’ and it gave me a grotesquely deformed ‘girl’ with four arms and three mangled legs sitting beside her own deformed head. I asked for a ‘man in a tree,’ and it returned with a tree trunk that had a pair of legs popping out of it like they were branches. I quickly gave up on my Hugging Face.

After dabbling with some enchantingly named programs like ‘Deep Dream,’ ‘Art Breeder,’ ‘Night Cafe’ and a handful of others, I eventually focused on the most popular programs. Midjourney and DALL-E 2.

Portal To The New Now. Daniel Simons. Created in Midjourney.

Midjourney is renowned for its spectacular and awe-inspiring art. Unlike DALL-E, which is run from a website, or Stable Diffusion, which can be downloaded onto a local computer, Midjourney is housed on a Discord Server (a social media platform), which has enabled it to nurture one of the largest and most engaged Generative AI communities on the planet. It's not as easy to use as other programs, but it doesn’t take very long to master the basics.

DALL-E 2 was created by Open AI, the Elon Musk founded, and Microsoft-funded organisation that is also responsible for creating ChatGPT. The name DALL-E comes from combining the surrealist Spanish artist Salvador Dali and the animated Pixar robot WALL-E. The second iteration of this text-to-image generator, DALL-E2 was launched in April 2022 and the upgrade has led it to become one of the best image generators around.

Midjourney and DALL-E have their own unique strengths and weaknesses. Midjourney is fantastic at creating spectacular, mind-bending, art, but it’s not as great at photo-realistic images. David Holz, the founder, says the program is ‘optimised for beauty rather than realism.’

Circular fashion shoot under a psychedelic sky. Daniel Simons. Created in Midjourney.

As well as the fact that it’s almost impossible to create a ‘bad’ image in Midjourney, it also has a great ‘blend’ feature, where you can upload multiple photos and have the program combine them into remarkable works of unique art. A downside of Midjourney is that most of your work will appear in public news feeds, which means anyone can use or copy it.

DALL-E will typically churn out less spectacular results, but it is more attuned to your specific prompts. The program's two best unique features are ‘Outpainting’ which lets you continue and expand any image beyond its original borders, and ‘Generate Image,’ which lets you upload an image and have the program create infinite AI-generated variations.

Once I had selected my tools, it was time to go deep. I joined every LinkedIn, Facebook and Discord group and channel I could find. I watched hundreds of hours of Youtube tutorials, followed Reddit threads, read every guide and open-source community spreadsheet I could find, and signed up to way too many Medium pages and Substacks.

After playing with DALL-E, Midjourney, and Stable Diffusion, I decided to focus mostly on Midjourney, and used DALL-E for variations and outpainting.

Then it was time to take a journey into mastery…

"Midjourney and DALL-E have their own unique strengths and weaknesses. Midjourney is fantastic at creating spectacular, mind-bending, art, but it’s not as great at photo-realistic images. David Holz, the founder, says the program is ‘optimised for beauty rather than realism.’"

Chaos Incubates. Daniel Simons. Created in Midjourney.

Level 1: Baby Dreamer

When you log in to the Midjourney Discord Server, something doesn’t feel right. It’s like logging into Facebook to create a word document. The first thing you have to do is join a ‘#Newbie’ channel. Suddenly you’re greeted by a dizzying stream of images. Strangers’ fantasies are coming to life before your eyes: movie posters where all the actors are humanoid cats, a ninja in a dystopian cyberpunk city - one man has uploaded an image of his head and asked the bot to transmute it onto a posing bodybuilder. These are the dreams of the Newbies.

To create your masterpiece you order the Midjourney ‘bot’ to ‘/imagine.’ As soon as you’ve entered your incantations into the machine, it's instantly created and then instantly lost in the cascading newsfeed and you have to scroll frantically to find your work before it disappears into the abyss. When you finally locate it, you notice that three other people have already started making new iterations of your original design. There is no room for ego or ownership in the Midjourney. It's free love for all.

Baby as an old man surfing. Daniel Simons. Created in Midjourney.

The bot spits out a grid of four images. You can select the ones you like and upscale them to a larger size, or you can keep iterating different variations of any, or all, of the four images.

Having your ideas materialise within 50 seconds seduces you to test the limits of your creativity: A unicorn reading a spreadsheet? A planet made out of artichokes and melted cheese? A flying solar-powered car in the shape of a koala? In a way, the bot and the global hive of creators are training you to dig deeper into the recesses of your mind.

As you order the bot to grant your wishes and sift through the communities endless stream of fantasies, you can’t help but feel like you’re inside the mind of some sort of collective, cybernetic consciousness that is slowly waking up as it feasts on the dreams of dopamine addicts and novelty junkies.

..and within minutes you’re out of free credits.

Congratulations, you’re now an artist!

The City is Regenerative. Daniel Simons. Created in Midjourney.

Level 2: The Wordsmith

You’re still a tourist but now you don’t need to look at a guidebook every time you need directions to the bathroom. You’ve paid for a Midjourney subscription, so now you can ‘DM’ the bot, and you don’t need to hunt through the #Newbie channels to find your creations.

In Midjourney, any monkey with two fingers can type in a prompt that leads to a thrilling image, but if you want to create a work of art that resembles an original concept, you’ll need a bit more sophistication. The craft of prompting is so vital that it has given birth to ‘promptologists,’ ‘prompt engineers’ and ‘prompt marketplaces’.

To master the art of the prompt you need to understand structure and syntax: first, you’ll describe your main subject. It might be a rooftop farm or a Tesla pulling a tiny house, or perhaps a magical creature made out of mushrooms and seaweed. Then you key in the surroundings, for example, a cityscape, a jungle or a desert. Next, you describe the form of the artwork, which could be anything from an oil painting, to a 3D render, or even street art graffiti. Finally, you’ll input the stylistic elements like the art genre or the artist you want the work to be inspired by - being conscious of the ethics of stealing a living artist's style, of course.

Regenerative utopian future city. Daniel Simons. Created in Midjourney.

Now instead of writing ‘A man in a tree,’ you’re writing ‘a man, standing in a tree in a lush Amazonian rainforest. Watercolor canvas. In the style of Dada.’ Instead of writing ‘woman in the park reading,’ you’re typing in '30-year-old woman reading a book in an autumn park in France. Oil painting in the style of Monet and Rembrandt.’

You can also have a lot of fun combining multiple genres and artists styles. In Midjourney you’ll finish your prompt with information about the parameters, like the aspect ratio and image quality.

If it's all too much to get your head around, you can use websites like Playgroundai.com or Prompter to build your prompts with templates and visual aids.

Level 3: Art Historian

After you’ve mastered the anatomy of a good prompt, the real fun begins. You can now ride the bike with your hands off the handlebars, but you still need to know where you are going. To speak the language of art, you have to become an art historian.

If you’ve joined the right communities you’ll be overflowing the spreadsheets and Github pages that go into detail about the thousands of different artists and art styles that you knew existed but never had a name for. Google also has a great dedicated ‘Arts Movements section’ in its Arts and Culture collection.

As you’re becoming an art historian, you’ll also dig even deeper into the specificity of each art form. If you want a photograph, what type of film, what lens, what lighting? If you want an oil painting, what type of paint? You’ll also start to master the more advanced parameters like ‘quality,’ and ‘seeds.'

Now you’re really cooking. Instead of writing ‘A goldfish at the beach,’ you’re writing ‘A goldfish, swimming in a glistening blue ocean. Photorealistic. 50mm lens, Fuji film. Cinematic lighting. Golden hour. Fauvism style. Auditory Synesthesia Style. Existential. By Beeple --ar 16:9 --quality 4.’

“After you’ve mastered the anatomy of a good prompt, the real fun begins. You can now ride the bike with your hands off the handlebars, but you still need to know where you are going. To speak the language of art, you have to become an art historian.”

“Even though the court cases are currently deciding the future of these new world-changing technologies, there is no going back.”

Level 4 - Program Interplay and Parameter Mastery

Now that you’ve mastered the art of the prompt and become a genuine art historian the next step is to use the strengths of each program to create magic. Start your creation in Midjourney and then import it into Dall-E’s outpainting, or upload it and generate variations. Create images in Dall-E and ‘blend’ them in Midjourney. If you're lacking inspiration, you can try using a prompt generator, or type your prompt into ChatGPT and ask it to come up with some more elaborate alternatives.

This is also where you might delve even deeper into Midjourney's more advanced parameters like ‘test,’ ‘chaos,’ and even ‘video.’

If you really know what you are doing, you can train Stable Diffusion on your own data sets, or export your creations, use an AI upscaler and then continue editing the finer details in programs like Photoshop.

The Future Has Arrived Early. Daniel Simons. Created in Midjourney.

As you continue to harness the power of Generative AI art, you feel like you’ve been endowed with godlike powers.

As you swell with wonder and try to fathom how this magic is possible, you also feel the vertigo-inducing thrill of knowing that this is only the beginning.

Even though the court cases are currently deciding the future of these new world-changing technologies, there is no going back.

We have entered a new era, where, as Jason Silva put it, ‘we are all becoming imagineers, limited only by our dreams.’