Adventures in AI with DALL-E
Recently, I came across a link to Image Creator from Microsoft Bing. It uses OpenAI's DALL-E to generate images from a text prompt. I decided to try it out since it is free to use, at least for one hundred images or so. If you go directly to the OpenAI website, they want you to spend $15 to purchase credits—I'm not sure how many images you can generate for that amount. Anyway, free is free, so I tried my hand at generating Allison and her squad from my YA fantasy series, The Allison Lee Chronicles.
Allison by DALL-E
The first image I created was of Allison. I think the image works pretty well on two levels: it captures her appearance and snarky attitude. In the novels, she's described as a biracial teenager with forest green hair. Whenever I put forest green into the prompt, I ended up with a woman with green hair in a forest. For this image, I used green. My prompt was "biracial (Chinese and Caucasian) young woman with green hair". Adding teen or teenager to my prompt with any reference to Chinese or Asian generated a content warning and did not produce an image. Interestingly, if references to race were removed, I could generate images using "female teen" or "female teenager" as part of the prompt.
Another interesting point is that DALL-E often adds a nose ring and other piercings to the image whenever young is included in the prompt, even when I explicitly added no piercings (another character, Dalia, is described as having a gold nose ring). I actually had to edit the image above to remove the nose ring. Overall, this one fits my vision of Allison. To get closer to 100% perfect, she needs to have a darker shade of green in her hair, which should be a little bit longer.
Tough Dalia anime style by DALL-E
Here is Dalia in anime style. Hers was the easiest to create, probably because I didn't include racial adjectives in the prompt. However, I could not get an image with a satisfactory nose ring, so I gave up. The prompt was "tough teenage girl with neon pink hair anime style". Overall, I'm pleased with the result, as the image captures her neon pink hair and toughness. The biggest improvement would be a small golden hoop nose ring in her right nostril.
Haji cartoon style by DALL-E
Haji was probably the hardest to generate. I only got something close to my image of him by adding cartoon style to the prompt. He is described as a gangly boy with dark wavy hair and a son of Pakistani immigrants. Putting any reference to Pakistani or Middle East in the prompt generated images of super fit, handsome guys with sculpted beards—even when teenager or youth was included in the prompt text. Finally, I added cartoon style, and this was the result. I think it's great. The image captures his appearance in terms of wavy hair and slender build and hints at his goofy personality.
In the end, using DALL-E to generate images of Allison and her squad was a fun exercise that took little time. I spent probably 40 minutes or so to generate the three images above. At some point, I'll try my hand at generating images for a few of the antagonistic characters in The Allison Lee Chronicles.
Was this a helpful exercise in terms of my writing or creativity? Not really. Maybe. I don't know. It was entertaining enough, didn't feel like a waste of time, and provided content for this blog post. Have you tried an image-generating AI? Let me know about your experience in the comments!