AI Masterpieces and Mishaps: "make it so" by Elizabeth Tybush

Happy Pride!

Today’s blog is a little treat, brought to you by a chaos flare-up I had last night. I wanted to see if AI would produce any Star Trek: The Next Generation content if I simply used the prompt “make it so” with no other text (no genre, no colors, nothing!). DALL-E 2 produced four separate sample images, while Midjourney produced a single image with four options collapsed into a single file.

An abstract, surrealist image of what is probably a band, dressed in suits. Three of the four wear hats, all have a white complexion, and all appear to have red-orange hair. Music notes--or what could be interpreted as them--float against a black background. Two figures hold instruments that may be a violin and a Fender Stratocaster. One of the band members seems to have their face wrapped, almost like the Invisible Man, and wears a top hat with blue goggles.Generated from the simple prompt "make it so" with DALL-E 2, May 31, 2023.

An abstract, photorealistic image of a female figure with long, thick blond hair and a white complexion beside a grass-bordered pond. The figure wears white capri pants and a white tank top, and appears to be dancing ballet. There is a tree line yonder. Generated from the simple prompt "make it so" with DALL-E 2, May 31, 2023.

An abstracted figure, masculine and white complexioned, wears a dark suit and hat, and holds a guitar that looks eerily similar to a Fender Telecaster, but somehow, also a Gibson Les Paul. Fire sparks from where the neck and body of the guitar, but it's smooth and magical, and whirls up into the star, galaxy-like sky that is also the backdrop for the entire image. Generated from the simple prompt "make it so" with DALL-E 2, May 31, 2023.

Two oddly realistic but definitely abstract people stand on a snowy city street. An adult man with white skin and short brown hair but also a bun stands a head taller than the woman of color beside him. She has brown skin, and her black hair is styled into an abstracted oval shape, almost 1980s goth vibes. They both look to the right of the frame at something off canvas. Generated from the simple prompt "make it so" with DALL-E 2, May 31, 2023.

A set of four square images; this is the preview image produced by Midjourney AI when given prompts, which produces images numbered 1-4, read left to right, top to bottom. Weirdly, these images stacked with similar colors. Images 1 and 3 have a soft pink-peach background, while 2 and 4 have a sea-green background (although 2's is detailed and contains a sky and sea). Nearly every image produced appears to have a third dimension, not online video games or Pixar-style animations. The styles are eerily similar, and yet while there are details between each that are the same, it is unfathomable how this occurred. Cloud-like shapes or outright clouds exist within a candy-colored palette. And unusually, while images 1-3 depict buildings of a fantastical nature, image 4 depicts a female figure with their head in the literal clouds. It is a bizarre collection. Generated from the simple prompt "make it so" with Midjourney AI, May 31, 2023.

I have no idea where any of this came from. I tried to do an image search (using Bing) with the same parameters, and it was everything I expected: Captain Picard. Captain Picard everywhere. Mostly things like…

Captain Jean-Luc Picard: “Make it so.”

The thing that bugs me is that each AI seemed to generate images that were related to each other in their own way. DALL-E created what I’m choosing to call (and maybe others have or this is something already, whatever) abstract photorealism, and 3 of 4 images have some relationship to music. While Midjourney seems to pull almost exclusively from digital art that lives in the uncanny valley, and weirdly chose from similar color palettes, focused on centralized subjects (with 3 of 4 being fantastical buildings), and contained clouds or cloud-forms.

(Writing the alt-text for all of these images was wild…I have no idea if I did any of them justice because the purpose of sharing these is not just for the basic descriptions, but to capture the chaos and the relationships between the images! Please get in touch if the gallery carousel is overriding that alt-text, and I will try to find a fix.)

I do often wonder if Midjourney, for as lovely as its generations can be, has either limited training materials, or if the community (which is in a Discord bubble) is sort of collectively copying each other’s prompts, giving Midjourney its own “personality,” so to speak, because the community also can rate and report and offer feedback and such. I do not know enough about how it learns from user input to say if that speculation has any merit, but it’s interesting to think about. At some point, I will attempt to share the collection of Midjourney generations that all seemed to have the same signature looks, no matter how the prompts were engineered, to avoid certain productions that continued to pop up. A massive undertaking, but one I hope I can find the time to put together!

Either way, I was amused and decided to share the chaos with all of you. Until next time!

Elizabeth Tybush

AI Masterpieces and Mishaps: “make it so”

Comment