I used Stable Diffusion's AI to generate many of these images. Although AI can refer to billions of images from the entire world for guidance, it seems to lean towards imagery and styles from popular culture, unless trained out of it. Cities often look like either Edo or Tokyo or New York skyscrapers. If you request "a female elf with armor and a sword," you'll get characters that resemble Disney's Mulan. If you simply request a “mermaid,” you’ll get recognizable versions of Disney’s Little Mermaid. I'd like to avoid getting a "cease and desist" letter from their lawyers.
Training AI is like negotiating with a genii. It will do its best to give you what you want, but you have to provide very precise instructions. I quickly discovered that AI does not understand similes. In your written prompt, you can’t say “skin like a fish,” you have to say, “skin with orange scales.” To create original mer-people, I had to describe each major body part, choose the best of dozens of results, and with each iteration, adjust the description:
“Female (or male) human face, orange hair, red eyes; human torso merging with fish tail, orange scales, fins. Full figure, swimming, side view. In the background is a kelp forest in blue water, with rays of light. In the foreground are pink shells and red coral. Full image.”
In addition, it almost seems like this AI is being raised and taught by programmers with the hormones and fantasies of 12 year-olds. Because it defaults to adding to large breasts, full lips, big butts, fangs and claws, dripping saliva, spikes, and lots of armor and weapons to every character. I have seen eye shadow on monsters. It gets confusing. Would you want these people raising your child? 🤣
Therefore, I also have to specify:
“NO teeth, no claws, no tongue, no legs, no armor, no breasts, no big butts, no clothes, no armor, no weapons, not scary.“
And yet, the version I worked with is also quite a prude. if I specify "no clothes," I have to confirm that I am over 18 years old. When I requested a golden helmet and "breastplate" on a character, I was required to change it to "chest plate."
As you might imagine, I am most interested in creating my own images and building my own characters. So I'm figuring out how to upload my original art work to AI, so that it can use it as a guide to building characters and environments. Eventually I'll be able to rotate, apply textures, add lighting, and more. It's very exciting! I’m reading about SD checkpoint models and LoRas. More to come.
One of hundreds of mermaids.
Copyright © 2018 Li Gardiner - All Rights Reserved.