With the posting of Chapter 12 and Chapter 13, we’ve reached the conclusion of Part One! Chapter 14 (next week) begins Part Two.
For this chapter, you’ll notice something different about the vibes image. Stylistically, it’s less impressionist and more realistic without being photographic. Over the past few days, I’ve been testing out different AI image generators (blog post on that sometime—not all are created equal). This image of Jemier was generated—after several failed attempts—by Midjourney AI. (Note: if you visit the site, be forewarned that is has an animation that plays in the banner, which is somewhat disorienting. If you scroll down immediately, you’ll notice an option to select an eye icon, which makes the banner stationary. Not sure why that’s not the default for accessibility.)
You might wonder why this failed. I hope to go into it a little more in depth in another blog post, but essentially, Midjourney I suspect has more inherent bias than DALL-E 2. While you can have extra control over your prompts and add “weights” to certain descriptors, it doesn’t always work.
Jemier, as you may already know, is what a character in future chapters calls a “beefcake.” And so to get those results from the AI, I had to think like the creators of any “beefcake” materials the AI was trained upon and use specific keywords. I continually got fantastic, vintage romance-novel cover results, except the following two things: he kept having blue eyes, despite my explicit ask for brown eyes (with various attempts to weight those as important), and he kept coming up shirtless, even when I clearly asked for clothing (I even asked twice in the same prompt). Even here, with me having asked for armor, you can tell this is just a mantle piece and that an extension of the canvas would reveal his bare chest.
And anytime you ask for a male with long hair, inevitably, there is wind blowing through it. Even when he is, as you’ll read in Chapter 13, napping on a couch.
So while I’m pleased with the quality of Midjourney, I am not pleased with the verbal and comp-sci gymnastics I have to do to get rid of bias. This image was ultimately brought into both Canva (it has a new magic eraser that fixed some things) and Photoshop to give him brown eyes.
Nevertheless, I was surprised to see something so sci-fi/fantasy and regal pop up, and so I’m calling this a vibes image of what Jemier’s royal portrait might look like.
Enjoy the new chapters, or use the below dropdown menu to catch up!
You must log in to post a comment.