The Big Idea
The image above was created as a sort of “cut scene” for my characters. I was working with this map of a mine I got from Patreon. I’ve gotten one or two maps from this creator and they were very helpful. The lowest floor of the map has this lagoon.
I wanted a dramatic reveal when they entered from the lower left (hopefully, which they did). So I asked the AI to create this lagoon and place the NPC I had planned there (a couatl trapped in crystal).
The AI Process
So i would love to say that creating the image above was as easy as telling ChatGPT what I wanted it to do. The truth is, though, it’s a little more involved than that. To create this image involved four things:
- About 6-7 prompts on ChatGPT to get the cavern
- A separate set of about 3-4 prompts to get the couatl in crystal
- Combining the two in Photoshop (using a few color effects on the couatl to blend better
- The Photoshop AI to fill in some of the water and add moss
Why So Many Steps?
There are a few things you should know about creating images using ChatGPT (and Dall-E).
With ChatGPT, there is not really making changes, more like starting over each time.
When you create an image and ask for a change, you are really asking for a whole new image that is close to what you had but with the new parameters involved. I first asked for a lagoon with a large rock and two smaller ones. This was the result.
And while it kind of fit the bill, I wanted something more dramatic. Eventually the prompt that yielded the image above was this, “Can you make it more of a first person point of view? Also, the smaller rocks are twice the size here and the larger one is surrounded by water on all sides.” It did have a few challenges (it gave me a random third rock in the center. So this is when I went on to Photoshop.
Photoshop AI Generator is Great for Small Changes
With the AI tool I was able to make some tweaks to the cave. Working with the Photoshop AI is very different, at least I think it is. I’ve found that less words are better. So I traced the third rock and simply put “water” to get it removed. I traced the rock surfaces around the edge and simply put “moss” to get it filled in. More words tends to add additional details you might not expect.
Some Blending Is Best Done By Hand
I had also requested this image of the couatl in the crystal.
The process of blending the two images seemed out of the realm of the AI (for now). I really liked the cave scene and this and I didn’t want to risk having them be reset. So I did some manual trimming, a few blends and a bit of transparency to put them together.
Final Thoughts on AI Use
It’s great. Really. Yes it can take a bit, and sometimes you don’t get exactly what you want, but for my purposes it works.
And to be clear, my purpose is not to replace a human. As you can see with the maps, getting them from a person was the better method (I’ve really not enjoyed trying to make maps with the AI so far). But for our private campaigns, having these images be striking and easy to create is a powerful tool.