Adventures in AI-Generated Art

Adventures in AI-Generated Art

AI-generated art is a hot topic these days. Color me intrigued.đŸ˜» So, I went on a bit of a quest to see what's happening. These are my adventures in AI-generated art in the last month. Yes, I should have been writing, but there it is.

NOTE: This is a graphics-heavy post with examples.

What Is AI-Generated Art?

At the risk of over-simplification, AI-generated art is created by a computer program that associates the mega-gazillion images on the internet with nearby keywords, then lets users create descriptive text prompts to have the program create images. The two most popular right now are DALL-E and MidJourney, but more are entering the marketplace almost daily. Most have free options right now, but offer the ability to buy credits and/or memberships. They aren't expensive to start, but costs can add up if you're the type to dive in and test all the possibilities.

Algorithms Are the Secret Sauce

From my not terribly systematic explorations, the difference between the various AIs is the editorial decisions behind the algorithms. MidJourney's images are intentionally pleasing. DALL-E is less interested in pretty, but abhors violence. I haven't played enough with newer AIs like StarryAI, Wombo, NightCafe, or Craiyon (formerly DALL-E Mini) to get a sense of their biases. This is important to know because it affects the outcome. Using the same set of keywords, otherwise known as a prompt, will get wildly different results from the various AIs. 

Bad prompts create odd images. I liken it to the old monkeys and typewriters theorem, where with enough monkeys on typewriters, you'll get a readable story. Similarly, with the right prompts, you'll get surprisingly good images. The trouble is, a good prompt at one AI is a terrible prompt at another. To complicate this, using the same prompt with the same service generates different images. This is intentional, meant to create unique images for each roll of the dice… er, algorithm. 

So while the AI generates the art, it's humans who come up with the prompts. Unsurprisingly, there's an emerging market in paying people to come up with prompts for you.

Adventures in AI Art

These are sample images from my various adventures in AI art. I've played the most with DALL‑E. I most recently tried Craiyon, but haven't yet figured out how to get images that are better than refrigerator art.

Wombo.art prompt: starship storm lightning comet red purple cyan galaxy alien (click thumbnail to see the full image)Art by Wombo.art from prompt by Carol Van Natta. Concept illustration of starships in a rain storm


StarryAI.com prompt: open book, worlds of magic, evolved using a previous image as a seed (click thumbnail to see the full image)

Image by StarryAI, prompt by Carol Van Natta. Abstract art of magical books


NightCafe.studio prompt: spaceship, high detail, HD, cinematic lighting (click thumbnail to see the full image) (Note: This is my newest adventure, so I've only generated one image so far.)

Art by NightCafe.studio, prompt by Carol Van Natta. Illustration of a spaceship.


MidJourney prompt: future space pirate fleet of starships attacking a big freighter spaceship, energy weapons, war, apocalyptic, octane render, cinematic lighting, –ar 9:16 (click thumbnail to see the full image)  (Note: This service is still in beta testing, and requires you to join a Discord server to enter prompts—see instructions.)

Art by MidJourney, prompts by Carol Van Natta. Starship battle in deep space.


DALL-E prompt: stained glass window of a pair of winged ocelots playing with a ball of yarn (click thumbnail to see the full image)

Art by DALL-E, prompts by Carol Van Natta. Stylized image of cats playing with yarn.

 

Wanna Play? Tips and Tricks for Your Own Adventures

If you want to play, here are some tips and tricks. 

  • Read the terms and licenses first. Some require you to attribute their service for any image generated by that service.
  • Don't expect photo-real human faces. Algos aren't that smart… yet. Plus, the services are understandably very wary of unscrupulous users creating deep fakes for profit or malice. 
  • Sadly, all the services exhibit sample bias and favor white-skinned, Caucasian, young, slender, and pretty when it comes to people in images.
  • Some services show you galleries and samples of images and the prompts used to create them. Look for images that catch your eye or are in the style you'd like to see, then pay attention to the prompts they used. 
  • Most of the services have forums, Discord servers, or at least FAQs to help you create successful images. Read them so you don't burn through the free credits with things that will fail miserably.
  • By using AI art, you'll be wading into an ethics storm whether you wanted to or not. See this article on a MidJourney-created image won a Colorado State Fair prize.
  • Fair warning: This is highly addictive for some people, by which I mean, people like me. Suddenly, the whole morning has gone by and you've spent all your credits for the day/week/month. 

Happy adventuring!


Bonus image from DALL-E, using this prompt: North American skunk wearing a kimono, attending a Japanese cherry blossom festival, ukiyo-e (I made this for my friend S.J. Pajonas, who wrote a science fiction romance story that includes a telepathic pet skunk.)

Art by DALL-E, prompt by Carol Van Natta. Japanese-style illustration of a kimono-wearing skunk at a cherry blossom festival.

 

1 Comment


  1. Ooooh! I love that you tried a bunch of different AI services here. I need to check out some more of these. And I still love that skunk image, It’s so cute.

    Reply

Leave a Reply

Your email address will not be published.