AI – Text to Image Generation Tools

One of the first things I explored in the realm of AI was text to image generation. It wasn’t my first interaction with this type of AI – I had checked out Firefly on my work account to generate a few things and even used Midjourney when it was first introduced. I wasn’t impressed really with either tool at the time and shelved the idea of using them for any real work. Two years later as I started really leaning into learning AI I decided to revisit the major text to image tools to see how far they’ve come along. To test them out I decided to feed them straight descriptions of some of the characters in the story/book I was writing. I created a prompt for each one and fed the identical prompt into each image generator.

The first thing I noticed was they really didn’t like making unattractive people. Even if the prompt specified that a person was older some models kicked out images of people in their 20’s. Even the old weathered warrior looked like he belonged on the cover of GQ. I know I could have tinkered with the prompt and really pushed it to the image I wanted but for this experience I wanted to use the straight descriptions from the book.

The tools I used for this were DallE3, MidJourney and Adobe Firefly. I know adobe is all about cutting edge design but their tool failed on almost every prompt. It ignored critical details in the prompt like ages, scarring, hairstyles, etc. and basically served up whatever it thought was close. Coming in second was DallE which tended to spit out anime leaning images (although to be fair, I did this before I learned how to craft hyper specific image prompts) and also couldn’t seem to handle more specific prompting. The clear winner was MidJourney – it came the closest on each character and the quality of it’s output was actually inspiring to me as it was how I viewed some of these characters in my head.

I know there’s other tools out there but I settled on the ‘big 3’ for this comparison so without further ado here’s the output from each tool for the major characters (so far)

After I got these results I did a deep dive into prompt crafting for images and got a lot better at directing the tools to produce what I wanted. I didn’t get a chance to go back and recreate all these characters but here’s some random images from the book that I created using these advanced prompts