How they are made
Our group of friends decided to start playing Dungeons & Dragons, using the Lost mine of Phandelver as our first adventure. Due to a (as it seems) chronically shortage of funds, we decided to test a couple of low budget forms of keeping track of combat. Using Roll20 for the Dungeon Master (me) it became a hassle to keep moving the players avatars across the screen. It was also not possible (due to space constraints, and again, budget constraints) for the players to keep their own positions during the game on laptops, so we switched to a so called Paper Mini setup.
- I am terrible at drawing stuff out of my head
- I want to make them to have an uniform style
I want to be able to generate all sorts of minis from images, not by painting them myself (see hurdle 1) while keeping the style uniform. A technique I was already using which accomplishes this task are Deep Dream generators. Deep Dreams are those unsettling images with dog faces all over them. They work by using Google's DeepMind AI for recognizing shapes, animals and persons in photos and images. An offshoot of Deep Dream is Deep Style where you supply the algorithm with an style image in addition to the source image. The algorithm will interpret the subject of the image, and "re-paint" it in the style of the style image.
This is epic because:
- I can use images as a "base" source and do not need to paint from my head.
- The styles are uniform as long as you supply the same style image for all the source images
For this Deep Style-ing there are a lot of tutorials to do it on a Linux PC (which in my case utterly failed because my old laptops graphics card wasn't nearly powerful enough), or websites. My favorite: DeepDreamGenerator.com. The site is intuitive, easy to use and if you spend some time making nice "dreams" your account gets a promotion and you can generate higher resolution images.
Finding the images
For the Deep Style to work I needed two types of images, a suitable style image and the source images of the monsters.
The easy part: finding the source images. Wizards (the company behind (among others) Dungeons & Dragons) has a lot of art (and I mean a lot!) on their website and on the DnD Beyond website. The awesome part is they allow for "fan content" to use their IP (intellectual property) as long as it follows a set of rules (read them all here). It basically comes down to this: I am allowed to steal all the images I want from DnD Beyond for use as my source images, as long as I don't charge anything (monetary or otherwise) for the end product (they allow donations tough).
The hard part: finding "style" images.
And down the rabbit hole I went...
The "art" of Deepstyle images is not in generating the images. Everyone can put two images in the generator and see the results. The difficulty is finding a style image and settings combination which not only generates a style you want for the images, but also an aesthetically pleasing one in stead of ugly splatters on a canvas. I found an image which gives the images the look I wanted after lots of trial and error.