Art Review: Stable Diffusion AI

I compare an AI artist to a human.

On an episode of The Sitch and Adam Show, guest Shadiversity discussed the potential of AI generated images. He noted how writers with no artistic talent could use AI to create book covers and thus forego the expense of hiring an artist. This got me thinking.

About two years ago I commissioned Caffeine Tweaker, or CT, to illustrate “The Left Hand of Fomalhaut.” She ultimately did the work for free, but I had intended to pay her. So, I wondered, what if I had used AI to generate those images? It way have saved me some money, but would they be as good?

Let’s find out.

It should be noted that I used a free web version called Stable Diffusion that can be found here. I also have no idea how to use this technology. My only instruction comes from the website’s statement, “Just enter your prompt and click the generate button. No code required to generate your image!”

I entered a variety of different prompts for the “Maggot Woman” villain who appears around halfway through my story. To begin with, I used the story text copied verbatim¹:

Most of its body was an enormous maggot, skin semi-translucent, innards visibly sloshing within. The mouth hooks had been surgically removed and there, attached at the waist, began the human female portion, emaciated and pale. The entire thing wriggled side to side in hideous locomotion - a sight made more grotesque by the exaggerated swing of pendulous breasts.

Here is what Caffeine Tweaker created:

Caffeine Tweaker’s version of the Maggot Woman.

And here is Stable Diffusion’s:

AI images generated by a verbatim copy from the story text.

Horrifying, to be sure, but not at all how I imagined her. CT came much closer.

Next, I used a single sentence that basically captured my conception of the monster: “Upper half of a woman surgically attached to an enormous maggot body.” Below are the results.

---

---

AI images generated by a single, explanatory sentence.

Once again horrifying, but not at all how I envisioned her.

I next tried a number of variations on the sentence like replacing “enormous” with “giant” and “maggot” with “larva.” I will not provide every single result but give a few of my favorites.

A random assortment of other attempts.

Finally, I read examples of prompts and discovered one can specify style with words such as “fantasy” or “epic.” Below is another assortment of attempts using key words like “Lovecraftian,” “Clark Ashton Smith,” and “horror.”

Another assortment utilizing "style" words.

This could go on forever. There are many variations one might try. Even the dozen or so examples here do not make up every image Stable Diffusion generated during this experiment.

Conclusion

There is no replacing the real thing. Yes, AI has created some incredible art, but one needs to know how to use it. With a human artist, on the other hand, there is no learning curve. In this case, I told CT what I wanted, and she produced almost exactly what I had imagined.

I will continue to play with this. I hope to generate decent book covers for my novels. I may also use it for future “Original Fiction” posts - once I complete some original fiction, that is.

---

1. Some text in the published version of “Left Hand” was changed to match CT’s depiction. That text was not used in this experiment.