In a previous article, we experimented with creating concept art through using text prompts to the Deep Dream artificial-intelligence image generator.
The challenge with this approach is that it can be difficult to specify, solely using text, the art you want – especially the pose and viewpoint.
In this article, we'll attempt to combine a text prompt with a base photo to generate concept art. The advantage of the base photo is that it shows the AI what you want.
![Editor Katie original photo Editor Katie original photo](workbench/pics/fan/2023/75425a.jpg)
For our base photo, our own Editor Katie
volunteered this photo from her beauty pageant hobby. A woman with wings! We should be able to generate some fantasy concept art from this.
From our previous experiments, we know that the Deep Dream AI 'improves' photos by referring to its vast database of art and pictures.
![Editor Katie improved Editor Katie improved](workbench/pics/fan/2023/75425b.jpg)
Above is an example of how the AI transforms the original photo, if given no other guidance. What we want to do now is to use the text prompt to control the transformation.
![Dragon Katie Dragon Katie](workbench/pics/fan/2023/75425c.jpg)
So we tell the AI we want a dragon using the base photo, and we use a medium filter strength in order to get significant changes from the original photo. (A low setting might not alter the photo much, and a high setting is less influenced by the base photo.)
![Dragon Katie Dragon Katie](workbench/pics/fan/2023/75425d.jpg)
It sometimes helps to add "in the style of..." to the text prompt. In this case, "in the style of Dungeons and Dragons".
![Dragon Katie Dragon Katie](workbench/pics/fan/2023/75425e.jpg)
And in this case, "in the style of Warhammer Fantasy".
![Dragon Katie Dragon Katie](workbench/pics/fan/2023/75425f.jpg)
You can also suggest a well-known artist – "in the style of Larry Elmore".
![Dragon Katie Dragon Katie](workbench/pics/fan/2023/75425g.jpg)
Or "in the style of Keith Parkinson".
Increasing the filter strength provides more potential, but more chance of failure. In the series below, the strength is increased gradually:
![Dragon Katie Dragon Katie](workbench/pics/fan/2023/75425h.jpg)
![Dragon Katie Dragon Katie](workbench/pics/fan/2023/75425i.jpg)
![Dragon Katie Dragon Katie](workbench/pics/fan/2023/75425j.jpg)
All of the fantasy pictures have been generated in Artistic AI mode. (There is a Fantasy AI mode, but it tends to produce animation-type results.) Now, let's switch to Cyberspace AI mode and attempt some sci-fi concept art:
![40K Katie 40K Katie](workbench/pics/fan/2023/75425k.jpg)
This is in the style of 40K.
![Space Ninja Katie Space Ninja Katie](workbench/pics/fan/2023/75425l.jpg)
Text prompt: "space ninja."
![Alien Katie Alien Katie](workbench/pics/fan/2023/75425o.jpg)
Text prompt: "female alien"
![Alien Katie Alien Katie](workbench/pics/fan/2023/75425m.jpg)
Text prompt: "insect-like alien"
![Mecha Katie Mecha Katie](workbench/pics/fan/2023/75425n.jpg)
Text prompt: "giant robot"
![Batwoman Katie Batwoman Katie](workbench/pics/fan/2023/75425p.jpg)
Text prompt: "batwoman"
![Superhero Katie Superhero Katie](workbench/pics/fan/2023/75425q.jpg)
Text prompt: "female superhero"
So that gives you some idea of what can be accomplished. Note that the AI has a strong tendency to stick with the original color range of the base photo. We also let the AI do its thing without a lot of guidance, and we haven't 'evolved' the photos (progressive revisions on Deep Dream). See how the results are influenced by the viewpoint of the base photo, and the pose of the original model.
(Also, I really should have cropped the base photo to just the figure portion.)
You might wish to use AI to generate concept art for future miniature lines, as we've tried here, or to provide illustrations for your wargaming adventures.