Meta-Prompting & Artificial Creativity
Last week I was invited to submit AI artwork for an upcoming expo titled “The Art of Equality: A Journey to Justice” hosted by the United Nations in Geneva in October 2023. I immediately thought of revisiting my portrait of Wangari Maathai, my first real commission in 2019 for New England Botanic Garden at Tower Hill.
Horticultural Heroes celebrates 20 plant advocates from the past and present, including figures who have sometimes been overlooked in history books. [S]taff paired artists from around the region with specific “heroes,” resulting in portraits in media ranging from pen and ink to acrylic paint to needlework.
I was paired with Maathai and the staff gave me a couple of licensed images to use as reference images. I used image style transfer or deep style to generate a few images that were composited in Adobe Photoshop. For “The Art of Equality” I used the older AI-generated image as a prompt in Midjourney, with a new feature called permutations.
Permutation Prompts allow you to quickly generate variations of a Prompt with a single /imagine command. By including lists of options separated with commas , within curly braces {} in your prompt, you can create multiple versions of a prompt with different combinations of those options.
Each time I ran my permutations prompt I got 4 x 4 thumbnails or twenty-four thumbnail images. In the curly braces I used art terms such as allegorical, cloisonnism, and synthetism. Cloisonnism describes post-Impressionist paintings with bold and flat shapes, separated by dark contours (think Paul Gauguin). Synthetism is derived from the French verb synthétiser (to synthesize or to combine so as to form a new, complex product).
In my book Techno-Vernacular Creativity and Innovation I refer to the term synthesize as a part of reconception, which engages improvisation and repetition. Professor William Hsu offered his take on my process to recreate the 2019 Wangari Maathai portrait using Midjourney:
If you compare the 2019 image on the left to the real photograph that was used as an image prompt, it doesn’t change much beyond the light model and background. While both of the artist’s creations look hand-constructed, an algorithmic inpainting shader could do the same thing with no machine learning; it’s mainly a work-saver.
The right-hand image does all of that and puts the subject (the late Kenyan political activist Wangari Maathai) in a never-before-seen pose, with a generated plant, different clothing and jewelry in a similar style, prompted by Nettrice. This not only represents creativity in combined image-and-text prompting — it incorporates generative art as a style. I think the image looks somewhat different from the real Dr. Maathai, but she is still recognizable, and the new image has both hand-painted and photorealistic aspects.
Hsu refers to the process I used as meta-prompting. For the UN expo, I imagined the new Maathai portrait as the center of a triptych — a picture on three panels — with Kenyan nature scenes on either side. I used the same text prompt without the Maathai references for consistency. I also edited Maathai’s face to make her more recognizable in the image. Shine (specular reflection) was added to the skin, as well.
There is a misunderstanding among AI art nay-sayers that the machine does all of the work in generating these images. I’m here to tell them that the human (me) is guiding the process from beginning to end, using features such as permutations as tools to create AI art.