It’s official, my mind is blown. We’ve been hearing about the art of developing a great prompt when working with generative AI platforms like DALLE-2 or Midjourney. We’ve been watching artists make amazing images and wondering how they did it, having no answers from them, and wondering if that mystery becomes part of their art.
We’ve also been watching artists create images while making their process completely available - showing us what they are feeding into the AI, and charting the iterations of what it produces.
And there have been early tools and guides created by enthusiasts to initiate exchange and knowledge about what prompts produce what effects - essentially geeking out on the sharing aspect.
(Midjourney itself only releases outputs through their discord so that everyone’s creations are public, along with the prompts they used to make them.)
But today I learned of PromptBase, which is a marketplace for prompts. At PromptBase you can buy (or sell) prompts that have been developed to create a certain output, and these outputs have hit a level of sophistication that is seriously notable.
The marketplace sells prompts for Midjourney, DALLE-2, and GPT-3.
An example of a GPT-3 prompt is Prompt Milker, which is yours for $4.99. Prompt Milker will “allow you to get much better answers than you would normally do, by arranging, or completely restructuring your request, adding more details and fixing errors.”
An example: Instead of entering “write a poem about the importance of money in our daily life” into GPT-3, Prompt Milker will tell you to revise your input to be “Compose a lyrical ode to the power of money in our modern world. Utilize a poetic style that captures the nuances of the topic, exploring the way it shapes our lives, our decisions, and our relationships. Consider the ways money can be both a positive and a negative force in our lives and craft your verse to reflect the complexities of the subject.”
First, let’s notice that this is a prompt that revises your prompt.
Second, the linguistic prompt becomes the interface to the AI. Maybe that was always true but somehow I really get the difference when I see this.
Third, let’s notice that the creator of Prompt Milker is drawing on descriptive talents that are based on some knowledge of poetry and a kind of sophistication that - dare I say it - might mean they studied the humanities. (Go team!)
Which brings me to the huge array of prompts available on this platform that are geared toward the text to image systems.
Here are prompts like 3D Abstract, Embroidered Illustrations, or Synthwave Nature Landscapes. For the moment, disregard these actual prompts or whether you think they are sophisticated or interesting. What I find notable here is that someone is creating a prompt that ultimately requires the kind of knowledge you would get in art or design education: knowledge about art history, visual culture, maybe technical aspects of image creation like specific lenses or depths of field, maybe descriptors relating to specific features in software tools like Photoshop or Cinema 4D, maybe an understanding of design history, architectural terms, or specific materials or techniques.
So a successful prompt creator would need to have deeper knowledge of (let’s say) a historical period than other people, and be able to conjure the specific terms to describe in detail the qualities of the desired image output. They’d need to be able to work from knowledge about the medium or tradition, bring it into language, and then use that language as a code that the machine can successfully understand. Those are art people!
These prompts remind me of apps. Will we see an artist create a runaway bestselling prompt that creates an empire?
As an aside, I wonder if this approach to designing and selling specialized prompts becomes the kind of monetizable layer that artists who are angry about copyright issues would be able to live with. Thinking of examples like Weird Album Covers, where an artist is essentially selling the ability for a person to create images in their personal artistic style.
Or where the plinth designers among us can make their fortunes.
By the way we may ask ourselves whether it is wise to use Midjourney to make a website.
PS: On the grotesquerie of humankind
This is also a shortcut to the infinite multiplication of predictably annoying stereotypes such as Sexy Heroes and Villains (only $6.99) and inane ideas such as Countries Re-imagined as Women in which you can achieve an image of a buxom, seductively mercurial woman who just screams “FRANCE” and who is of course young, conventionally attractive, white (and standing in front of the Eiffel Tower).
Enable 3rd party cookies or use another browser
Indeed. Mind blown.
At a much more elementary level, I've been considering the nature of creativity and wondering if AI equals the human mind when it comes to actual creativity. Everything I've read to date indicates there is really no difference. Both canvas the available spheres of knowledge and experience to create something new. And both are limited (or unlimited) by what is or has been available to them.