With all the hype around chat gpt-3 and AI image generation such as DALLE and MidJourney, I asked myself what would happen if I combined the two technologies? With thousands of people blogging, and tweeting about this topic I'm sure this is not an original question. My spin on the question though was this, how good of a kids book could I write and illustrate using these tools in one weekend, and with zero experience with writing or publishing books of any type? In other words my question is how good of a product could I make with minimal effort, talent, and skill in one weekend?
TLDR; I wrote and illustrated a kids book entirely using AI that you can now buy on Amazon.
To start this journey off I type in the following prompt to chat-gpt3. I show a portion of the response below.
Honestly it was pretty good right off the bat. One thing I didn't like is that there was no arc to the story. The only conflict was that Timmy dreamed of something more than being with his family. Which felt pretty rude, to be honest. So I made a small edit to “… but what happens after Christmas?” With that minor change, the Timmy's arc is now: alone in a forest, being picked as the center of attention for a family, wondering what will happen after Christmas, and then ultimately finding a permanent home in the north pole with the help of Santa.
Now the only thing missing is the illustrations. With MidJourney you provide prompts of what you’d like to see. My first attempts were less than spectacular.
This image came from asking MidJourney to produce an image of Santa Claus on his sleigh in front of a scrappy Christmas Tree. What is this? Why is Santa so creepy? What is going on with his sleigh? It kind of gets the general idea but it’s pretty terrible.
After wasting way too much time trying different prompts to get somewhat decent results, I caved and looked up some tutorials on the web. Within a few minutes I discovered my problem. MidJourney is on version 4 and by default it gives you results from version 1. Putting the --v 4 option now gave these much improved results when asking for santa in front of a christmas tree
Now this is more like it. It looks pretty incredible actually. The only thing weird about it is the hands… they look a bit too large for Santa’s body AND he has six fingers, but other than that perfection.
The last problem I had was how do you ensure that the style from illustration to illustration is consistent? One way to do this is to specify what style you want your image in. In my case I used “ in the style of Norman Rockwell” for all my prompts. This gave pretty reasonably consistent results, although sometimes the santa looks very different picture to picture.
Take a look at Santa's workshop imagined by MidJourney!
Again pretty incredible and in fairly cohesive style as the last one. But still some weird stuff is going on with the hands. It’s truly terrible at hands. Another weird thing is that this elf has a full sleeve for a tattoo for some reason… I don’t remember any Norm Rockwell paintings featuring that. I mean it’s interesting but for a kids book a bit off.
If you want to see the full book I've published it on Amazon. Or if you are nice to me I will send you a PDF. It’s not perfect. It’s not even great. But honestly it’s way better a kids book than it deserves to be.
Comments