Those of you with a passing knowledge of this blog know that I write on occasion. I haven’t been writing nearly enough, life just got in the way over the past couple of years. Doesn’t mean I don’t plan to get back to it (for a reasonable value of “soon”), or that I don’t seek out creative outlets.
With all the current buzz on AI Art, I thought I’d give it a poke to see how it works. Following are some images (whether ‘art’ or not) and some notes about what I think.
First I tried Stable Diffusion, mostly because it’s entirely free, without registration or anything (unlike DALL-E, which asks for a lot of info). Looking at inspiration, and because there’s discussions about using it for cover art, I just looked at the covers of my own novels for inspiration.
Fuffing about with variations of “hooded man facing Roman ruins on a misty night” got me the above image. There were a few retries, and that was about the best and most usable. So I thought to myself, “Cool, let’s try something more complex”. Ah, the naïveté of youth…
A prompt of “Hooded man riding a gryphon over the Colosseum on a stormy night” yielded the following:
The stuff of nightmares! Clearly even the machine thinks I should stay away from visual arts…
I played some more, but have to say that overall the results were very inconsistent, and not something I’d use anywhere.
I then tried Midjourney. At least registration was simple, even if it requires Discord which is like any IRC chat — if you don’t have ADHD, you soon will. One of the advantages though, is looking at what other people are using as prompt and what the results are. You can learn a lot by it, from how to imitate an artist’s styles to using different algorithms.
Both Stable Diffusion and Midjourney give you 4 results per prompt, but MJ also allows you to “continue” — make variations of one of the results, or ‘upscale’ it to add more detailed. So playing again with the theme of hooded man and Roman ruins, with added storms and in an oil painting style, I got these:
Kinda nice, don’t you think? (Click to enlarge)
Of course, I then just had to add ravens and gryphons…
I even tried to teach it what a gryphon (or griffin) is by describing the half-lion half-eagle parts:
Clearly it had an unfortunate encounter with a jet engine.
In no particular order:
I might use some of those for short stories, but my cover designer can sleep peacefully at night. You may get the occasional thing that kinda fits and can be used for an online blog, but for when you really need all the elements that make a consistently great cover across mediums (from eBook to Print), there’s still no match for conscious, directed design by a professional.
And that’s not even touching on the issue of copyrights. AI works by training models on data, images with associated descriptive words. Some of those images may have been copyrighted material, so even if you don’t use restricted images as prompts they art itself may still be exposed. This is early days without much case law, but still – there’s a raging debate about this subject. Even DeviantArt faced massive backlash after building an AI tool and training it on their users’ content. As an indie author you may not make enough money to be worth suing (probably will net more from being sued, what with all the free publicity of innocent artist oppressed by large corporations😉), but who needs this aggravation in their lives? So use with caution.
How much “intelligent” fits in the AI ART?
The ABC ran an excellent article about this: A journey inside our unimaginable future. It brings the story of Loab, a woman created with these AI models but constantly surrounded by disturbing images out of horror genre. The theme was hauntingly persistent across retries. The researchers even tried generating text to interview her (via text-based AI models).
Whether you want to see any deeper meaning in this or not, there is no doubt that these tools will have a dramatic impact on the future of what we consume as entertainment (or forced to consume as advertisement). It’s really is an excellent article, and I heartily recommend you read it.
It’s addictive – but is it art?
It’s certainly fun, and you get dragged into trying more variations to tweak and see what you can get. You can certainly burn hours on it (luckily, the Midjourney free credits run out at some point). But at the same time, even when looking at a great result, it isn’t nearly as satisfying as doing something yourself.
It’s too quick, there’s so little effort and therefore no sense of achievement. You just keep trying variations, getting meandering approximations. It’s like yelling at your child to do their homework (with about the same quality of results 😁). As my old teacher once said, good art is born from personal suffering.
If you do try
If you do want to burn an afternoon to see what’s the buzz is about, here are some notes to shortcut your learning curve:
Try Midjourney over Stable Diffusion (and let me know if DALL-E is worth the amount of personal details they are asking for — sorry if my dayjob is talking here).
Read the documentation, and observe what others are playing with (not just the glorious finished results, but the steps — and missteps — on the way).
Don’t get over complicated in the description, but use keywords (like how someone would tag an image, rather than describe it).
Try to imitate a particular artist’s style, if that’s what you’re after.
Add “–v 4” at the end, to use the newest algorithm which gives good results (better composition).
And if you like to sleep, avoid adding ravens!
As mentioned, Midjourney also allows you to use an image for inspiration. So using my author profile image, described as a ‘daydreaming author’, algorithm version 4, and contrasting the styles of Rene Magritte and Alphonse Mucha, yielded:
Help me get back to writing! Read the free short stories and novels, and leave reviews! I got the next volume, In Victrix, written and half-edited. As soon as I’m settled in the new place (oh, didn’t I mention? We’re moving interstate in December), I’ll get back to it…