The Butlerian Jihad used to be, for me, one of the most fantastical elements of the Dune series. Sure, melange enabling interstellar travel and giant worms, but a war fought to prevent anything like computers?

With the rise of generative AI models, I think I see it now. It isn’t that humans fight against the machines, it’s that humans fight against humans to prevent the use of the machines.

I now believe Dune to be among the most realistic science-fiction sagas ever written. It’s about a company strictly controlling a valuable resource that enables transportation, and being willing to do anything at all to preserve their monopoly.

The oil must flow.

Presenting without comment, AI model MidJourney v4’s response for the prompt “Black history month.”

My daughter texted me urgently today, asking for help in art class. As someone who attended school before the advent of cell phones, this entire concept threatens to derail me, but I’ll focus for her sake. She wants examples of abstract art, and wants me to help, and I quote, “Using your uhm ai art thing.”

I happen to have the Dream app installed on my phone, and haven’t used it very much, so I started there. That app heavily leans into models trained in different styles, one of which is “Abstract,” so it seemed like a good bet. For the first prompt, I supplied an input image, which I said should heavily influence the result. Avid fans of anime might guess from the result that the image was fan art from “Assassination Classroom,” which was also the text prompt. For the rest I supplied no starting image, using only text prompts to steer things. Unfortunately, I didn’t keep notes at the time, and worse, I foolishly didn’t “Finalize” any of the results, since I was just looking for samples to influence my daughter’s art on paper. So I don’t remember what words I used for the other three examples, my favorites of the eight I generated for her.

If I can generate image description alt text for abstract art, it should be easy for anyone to generate image descriptions for anything.

Since a previous post alerted me to the idea of an AI art model generating images of Buddha statues without having “Buddha” as an input, I decided to test it, so I prompted MidJourney with “cree, pasadena, ish, informal, preview, lindo, kids, potential, statue, calcutta, phenomenal, sigma, chero, heh, âľĵ, kier, bourdain, anjali, ori, displa”

The first time I forgot I still had the “anime” filter turned on, so I switched to MidJourney model v4 and re-ran it.

I’m not seeing Buddha in my results, although I do see statues. Then again “statue” is one of the words in the prompt, so nothing magical there.

I’m including both results because I guess I want to spend a long time writing image descriptions. 😀

This site presents “both sides” of at least one aspect of AI art by displaying the content of a current lawsuit over AI art on the left, and via fair use, a response or rebuttal on the right.

(Replace frivolous with litigation in the URL to get the original site.)

I’m obviously a fan of AI art, and occasionally post it here. I’m also very far from a copyright maximalist, although I’m not in favor of abolishing copyright altogether. I agree with the right side of the site on at least two points:

  1. Copyright law doesn’t apply here.
  2. The suit dramatically misrepresents, I think because Butterick doesn’t understand, how latent diffusion works.

That said, while I overall agree with the critique, I think I’m more sympathetic to the complaint than the rebuttal writers are.

There’s something truly disconcerting about adding an artist’s name to an existing prompt and seeing the result change to be much closer to something in line with previous works of that artist.

I can understand that statements about how latent diffusion works fall flat after seeing that. No, there are no images stored in the model, not even textual representations of images. And yet, there’s something, right? It’s not nothing. There’s some sort of minimal association of given names with a given set of characteristics, more so for some than for others, and the effect is often uncanny. The example of what seems like an arbitrary series of words consistently producing Buddha statues is a good start to counter that, but it doesn’t quite seem adequate.

Overall, I think there are serious ethical issues with how art is being sourced for training the current generations of AI art models. However, I don’t think there are legal issues; copyright certainly isn’t being violated.

More to the point: the primary users of these tools currently seem to be artists. In that sense, it really does seem like the advent of photographs, or the advent of photoshop, and it will likely be quite a while before we will know how these issues settle. In the meantime, anybody using these tools to try to mimic a particular artist is probably missing out.

I finally managed to get my daughter interested enough AI art enough to try some prompting of her own, and she immediately put Midjourney to the task of creating a new profile picture for one of her many online accounts.

On this fifth day of Christmas, I hope Santa is taking time for himself. Maybe relaxing, maybe getting some sun.

I asked Midjourney what that might look like.

I’ve posted a couple of Fiat-related things recently, so why not one more? I asked MidJourney to give me an image for “fiat 500e gran prix hyperrealistic” but it was four months ago, so I’m not exactly sure which version of the model was active at the time. This was the result.

I asked MidJourney’s v4 model to generate some images, then grabbed one of the four results for each prompt and upscaled it. Horrifying things happened to the number of fingers, and I would definitely need to keep working with this prompt to come up with something I would be willing to show in public as anything other than an AI art experiment.

Another pair inspired by @LeviKornelsen, also using a single added phrase: “in front of stained glass.” The MidJourney v4 prompt text is mine, based on descriptions Mr. Kornelsen gave his processed images.

I hope readers appreciate how much effort I’ve been putting into the descriptions for these images!

I find myself intrigued more and more by what just one added phrase can do to AI-generated art. In this case, the phrase is “in front of stained glass,” and this entire series was inspired by @LeviKornelsen

As usual, these were generated with the MidJourney v4 model, and the complete prompt text is in the image description.

When generating art with MidJourney, one of my favorite things to put in a prompt is “magic hour.” Here are some results with and without that phrase; see if you agree!

In both cases, this is MidJourney v4, and the only addition to the prompts was “, magic hour”

My daughter wants beef tacos for lunch, so I asked MidJourney for help.

The “beef tacos” are almost normal, but the “beef soft tacos” have unusual titles add for no obvious reason, and definitely no benefit.

Today I tried a few new prompts, no refinements, just simple phrases. These are from MidJourney, and my settings are: MJ version 4, Base quality, Style med, Regular upscale.

The v4 model often seems to add titles where none are needed or wanted, but overall I found all eight images pleasing. The waterfront pics don’t look like Hong Kong at all, but they’re all nice-looking waterfronts. The night market pics look more like Hong Kong, except all the Chinese writing is broken.

When I talk about “stuff” on women’s faces being produced by MidJourney, this is an example of what I mean. I’m not at all sure what’s happening with the figure on the left.

When I was reviewing images before the -v4 option launched, it was nearly every image featuring a woman’s face. It’s not quite as bad now, but it’s still marring far too many prompt attempts.

But hey, I don’t want to complain, so I’ll add a second image. This one actually had some facial marring when it first popped in one quadrant, but when I upscaled it, the face cleared up.

I tried the prompt “A girl with very long pigtails, eating a pear” in Idun.AI’s new generator using both the Idun.AI model (red fruit) and the StableDiffusion 1.5 model. The latter seems more photographic, but also uncanny, while the former will never be mistaken for something real, but works as an image. For me, at least.

I’ve had some challenges with MidJourney adding “stuff” to faces, mostly women’s faces, and prompting with “clear skin” doesn’t always help. Still, this result for “indian girl, suit and tie, photograph, 32K” turned out pretty nicely. It’s using the v4 model candidate.

Let it not be said that I’m whinging about mastodon without offering any solutions! Even if the solution is Yet Another Explainer, I guess I’ll yeet one out anyway.

What is mastodon? One way to think of it is as a mashup of Twitter, Discord, and email.

Like Twitter, people use it primarily for microblogging. People post short blurbs of text, or photos, and follow other people doing the same.

NOTE: You need not actually sign up for mastodon to see someone’s posts (assuming they’re public). It is perfectly reasonable to bookmark a favorite writer and just read.

Unlike Twitter, people aren’t all dependent on a single website run by a single company, owned by a single oligarch. Instead…

Like Discord, people are on different “servers,” or instances because we nerds don’t actually speak normal English when not required to. I’m currently on fifteen (15) different Discord servers, so when I want to talk about I switch to one of those servers, and when I want to talk about I switch to one of those servers, and so on. But…

Unlike Discord, you need not actually join each of fifteen different servers. You can visit other servers, and follow people without joining the server they’re on. Also, even though they use the term “servers,” Discord server are all still dependent on a single company behind the scenes.

If you’re familiar with Discord, you might be wondering, how does that work then? How do I go see what’s going on in the world of without visiting the Discord server dedicated to it? And that’s where the email comparison comes in.

Like email, you don’t need to share anything with another user to communicate with them. Maybe you use one of the big ones, GMail, Yahoo!Mail, or Hotmail. Maybe you use an email address provided by the telephone company that provides your internet connection. Maybe you use a work email address. It doesn’t matter, you can send and receive emails from any of the above, no problem. And yes, I’m glossing over some things there, but the point is, you pick your email service provider because of the user interface or the features or the price or because it’s forced upon you or because someoone else set it up for you. It doesn’t matter much.

And that’s key, and confusing! When you decide you want to join mastodon, it’s not like joining Twitter or joining Discord, it’s more like “joining email.” Because you don’t actually join “email.” Instead, you sign up for an email service. Chances are good that you sign up for GMail or Yahoo!Mail, because they’re free and easy and well-known.

So it is with mastodon. You need to sign up for a mastodon service. Don’t overthink it! If you sign up for one and later decide you wish you’d signed up for one with a cooler name or that several of your friends all ended up in one place, or you just learned there’s an instance dedicated to tabletop gaming that you’d have signed up for if you’d only known, that’s okay! Because…

Unlike email, when you switch servers, everything comes with you! Your followers are still following you and your posts are still there.

That’s enough explained for now. Mastodon isn’t super-smooth everywhere, but that’s part of its charm. It’s also not super-toxic unless you join a toxic server. So don’t do that! If you want to pick a server on your own, there are links at for that. But if the whole thing seems really overwhelming and confusing, let’s make it easy.

I’m going to list eleven servers, all of which as accepting new signups as I write this, and all of which are “general” servers, no particular focus or specialty. You can click on each to decide which one “speaks to you” most, or you can click on one at random and just sign up! I picked these manually as good general-interest servers, and then suffled them into artbirary order using so I’m showing no favoritsm. Just pick one!

And that’s my

Two more from the Midjourney v4 model:

“37-year-old man in a Hawaiian shirt, photograph, 32k”


“37-year-old bald man in a Hawaiian shirt, photograph, 32k”

Perhaps not as egregiously off-target as my previous runs at 50, 40, and 72, but they still make my brother look pretty good.

Midjourney v1.4 results for:

“72-year-old man wearing a sweater, photograph, 32k”

“47-year-old Hong Kong woman wearing a t-shirt, photograph, 32k”

“47-year-old Chinese woman wearing a t-shirt, photograph, 32k”

My father and wife both look younger and more attractive than these images, so if flattery is part of the latest model, it’s working.

Show more
Qoto Mastodon

QOTO: Question Others to Teach Ourselves
An inclusive, Academic Freedom, instance
All cultures welcome.
Hate speech and harassment strictly forbidden.