is not a real big fan of pre-painted minis, and he positively despises randomly-packed "collectable" ones - so why is he writing this article?
In my last article, we looked at making significant modifications to portrait-type photos.
But… could the artificial intelligence paint a mini?
This is a picture of an unpainted Vietnam War SEAL, as recently released by Gringo40s.
Since the artificial-intelligence model is trained on a wide variety of art and photos, does it understand enough to 'paint' a Vietnam War SEAL?
I set the filter at the lowest setting (10%), PhotoReal AI model, and specified "Vietnam War SEAL in color." It enlarged the photo, and to my surprise made significant changes to the headgear, gun, and figure stance!
At a higher filter setting, the result becomes more like a photograph of someone, but still no color.
Switched to Artistic AI model, and asked for a result in the style of Dan Nance. Specified boonie hat and Stoner 63.
I turned the filter strength all the way up, and finally got color! But the figure is complete fantasy.
Giving up, I used the Cyberspace AI model, and asked for something in the style of 40K. The proportions look off, but the AI came close on the basing! Apparently, it knows 40K!
So what we've learned is that to get the AI model to 'paint' a miniature, you have to turn the filter strength to such high settings that the result looks nothing like the original mini. And that the AI model knows very little about the Vietnam War.