We’re finding that, when it comes to card illustration visual themes, generative AI tools like Stable Diffusion help but they’re no silver bullet. For example, take a look at these technically suitable-ish images of angels for our trading card game. Some of them look detailed and recognizably humanoid (and some of them are outright awful looking). However, the juxtaposition of fantasy armor with busy urban sidewalks and streets just looks weird:
So, obviously some of these images are just botched ai output. Others have heads awkwardly cropped out–the whole list of usual image generation errors.
Even so, AI image generation proves to be a seductive trap for more technically inclined folks like me. Stable Diffusion’s output can only be as good as my input prompts. Garbage in garbage out. Quality in –> some garbage out and a few home runs.
The garbage in this case is the experiment mixing fantasy armor with modern urban surroundings. We’ll keep at it. Every once in awhile we get something really nice with the right combination of prompt, model, parameters, and visual theme:
