MidJourney

I have been trying to get MidJourney to create images that aren't a Lovecraftian Cthulu-scape.  The program works well for poems and abstract images of static objects, like buildings and trees.  I had fairly good success putting in lyrics from an old We Three song the other day ("...the moon sails over city lights, sails over stars and the water's edge.").

MidJouney did fairly well with still life prompts, like "High tea in a garden with roses and lavender," but not so well with "Black cat having high tea in a garden with roses and lavender" until I also pointed it at a picture of our cat, Cicero.  Still... the AI has a penchant for adding third (or fourth or fifth) eyes, drawing cyclops-cats, or inserting demonic felines.  Or just making lopsided kitties.  One particular specimen looked like the love-child of a cat and a snowy owl.

I am reminded of a 1970's Sesame Street short where a disembodied man's voice is asking disembodied children's voices how to draw an elephant, which has many moments of kids shrieking, "Not like THAT!" and laughing. 

Story Images

I've tricked out the Secret Writing Lair with a large leaded crystal, and now little rainbows are splashing themselves over the grape leaves and the foliate head.  This is a good thing, as it keeps my my mind off of the rising temperature.  At some point over the next week or two, the grapes will have produced bunches that are sweet enough to eat; we're hoping to get to them before resident night critters do.

Thursday night's full moon was more a meditation than an actual ritual.  I lit some incense, sat in the backyard circle, and listened to the evening sounds of frogs and crickets and the fountain.  Shortly after the moon rose, some neighbors about a block up the street began howling (later, Mark called it a full moon party).  It was kind of cute, but struck me at the time as slightly performative.  

I've continued to play around with MidJourney.  It's great for rendering abstract or general images, but I'm finding it difficult to direct the AI to produce images with a specific narrative that could be used as e-pub cover art.  Either the images become more uncanny as the AI adds extra heads, limbs, or body orifices or else it latches onto rendering an irrelevant object.  An attempt at a Victorian man at a large fireplace holding a chess knight resulted in a horse-headed gentleman with a hearth in his torso.  I think the trick is to not get too carried away by the AI's variations, which tend to become more bizarre and surreal as one goes along.

So, if I wanted an image that  Leonora Carrington might paint, the AI works great; aliens, I'm covered; specific fantasy story characters... well, if they were on fire or undead or extras from The Nightmare Before Christmas, I'd be good.  I did have some success pointing the AI to images to use as a starting point—Capricorn renderings looked more like goats and less like early experiments by Frankenstein.   Creating an image that will convince a reader to read the story, well... that's going to take some extra work.

Practically speaking, there might be a legal issue using AI generated images as cover art, because an AI (or the person using it) can't hold copyright for the image.  I suppose I'll have to look into the difference between using generated images and using public-domain images.

Writing wise... I reviewed some short stories and polished some bits that are more obviously in need of it after a period of not looking.  Another manuscript I looked at either needs to have the character's stakes heightened and the word-count shortened, or else I need to sit back and do some very loose outlining because the manuscript right now feels like an introductory chapter to a longer story.  I can see that my penchant for visual description needs to be reigned in, too,  

And I need to gear up submissions, too.

MidJourney Near Misses

Well... I've spent more time on MidJourney trying to get an image of a character that doesn't look like he's doing a dark rite by the flame of his orange eyes than I expected.  Since the character is gay, I added "gay" and "romance" to the end of the text and the AI responded the way I thought it might by rendering some shirtless eldritch horrors—and some eldritch horrors with shirts made of the skins of buff Harlequin Romance male models.

I'm using text from a story, and for whatever reason the Dark Fantasy switch must be on.  The text is a quick description of spell casting, although it's more of a charm involving jack-o-lanterns than a full-blown spell.  Maybe the Halloween vibe is too strong.

Oh well. 
 

I had better luck with an image loosely based on iron pentagon of Mars.

Continuing MidJourney Adventures

I'm discovering that the MidJourney AI has not been trained in classical Greek and Roman mythology, and that it's more likely to pick up modern cultural references to stories rather than the original myths and legends.  It also doesn't seem to know how to render Capricorn, the sea-goat, and instead of producing an image of a goat with the hind-quarters of a sinuous fish, it's more likely to make encephalitic creatures with crab feet and mutant horns.  I suppose I have to learn how to weight descriptors and choose the right ones.  

Also, if you ask the AI to draw librarians, it will render them as women.  Which does raise the question, why did I choose to have the librarian characters be male in the story.

I'm not sure if I'm having better luck using excerpts from various short stories and seeing what the AI does.  It looks like the AI wants to illustrate a Dark Phantasy, because as I cycle through the prompts, people's faces get more lopsided or sinister looking, and objects mutate into molten, multi-eyed parodies.  I might have to tack upbeat words like "wholesome" or "wonder" to the end to keep the artwork light.

It's also possible that my stories are creepier than I realize. 

So far, this has felt like homing in on a specific character in the multiverse.  Sometimes I get closer to the image of the character, and other times the AI goes in a different direction.   I'm not completely sure, but it's feels more satisfying to have another human come up with story images because there's a real sense of connection via an idea when someone else's art resonates with what was in my head when I was writing.  

Playing with MidJourney

The latest distraction fun thing I've discovered is the art-generating AI bot "MidJouney."  This has required that I also learn Discord, which I suppose isn't too bad a thing. 

I've been playing around with it over the weekend; the justification for doing so is that I can use it to generate covers for some short stories I've got lying around that I really should publish. 

Working with the AI is interesting.    I've managed to create a few images in the blended styles of  Edward Burne-Jones and William Holman Hunt.  I asked it to render tarot cards in the style of Pamela Coleman Smith, and what came out was interesting, but looked more like Crowley's Thoth Deck and a possible portrait of Ms. Smith.  

The AI doesn't appeared to have been trained on Middle Kingdom hieroglyphs, and I'm thinking that its renderings of the ancient Egyptian god Thoth may be pulling more from the Stargate Science Fiction franchise than actual ancient Egyptian images. I might have to try something specific like "man with an ibis head." (which is has led to depictions which are closer to either to Native American Raven or a medieval plague doctor).

I did a fine job with a flying machine over a medieval city.  It makes pleasing images of the moon and architecture, and I got some interesting portraits with the prompt "The Magician Tarot Card by Pamela Coleman Smith."  MidJourney does have a tendency to blend things together into one image unless one is specific—so far my attempts to have the Lord of the Animals and the Greenman dancing have resulted in a chthonic figure in the uncanny valley; "dung beetle hieroglyphs" resulted in images of a beetle with tiny pseudo hieroglyphs on its body.

There's some commands I can embed in the text that I tell MidJourney to build a picture from, and it looks like I'll have to play around with them some more.  

Er.  I mean.  I'll need to enter in some text from some of my stories and see if it comes up with some good cover art.  Yeah.