We've recently been corresponding with an English teacher in Alberta who took us up on our offer of custom images based on student requests.
The result is the Alberta Request Line image theme in Frankenstories.
To our delight, the class sent us a pack of material about their daily lives—descriptions of herding cattle, feeding ducks, playing sports, plus snapshots of the environment and ideas for various adventures and hijinks.
Using that inspiration, we spun up a pile of images for the students to curate and publish as a theme for everyone to play with.
This morning we got to meet the class over video chat and play a game with them to take the new theme for its first drive, resulting in this moo-stery about some missing cows and an ancient curse.
It was a great experience for us! We play a lot with our team, but the dynamics of larger groups and younger players are always different, so we're always happy to take the opportunity to play with a class. So much energy! We all needed a lie down afterwards, and not just because of the 6am start.
And the new image theme works great! So if your class wants a blast of frosty Canadian country air or likes farm animals gone rogue, check out the Alberta Request Line! 🐮
How we made the images
After our game with the kids in Alberta, we had time to answer some questions.
Obviously, the first questions were about wrestling crocodiles and boxing kangaroos (I had to disappoint everyone by saying our biggest issue in Brisbane was stopping dogs from licking poisonous toads).
But after that, everyone wanted to know how we made the images, so I thought I'd share a little of that conversation here.
I've written before about how we use DALLE and Midjourney, but our approach changes as the models themselves change.
Currently, our first pass is usually in DALLE via ChatGPT, because, while DALLE is limited stylistically, it produces the most coherent images for the kind of narrative scenes we like to use in Frankenstories.
If you want a kid riding a dirtbike while being chased by an angry moose, DALLE will give you a very DALLE-looking image, but it will probably make sense.
In contrast, Midjourney is just as likely to give you a boy with antlers riding a moose on wheels.
Also, and this is something I really appreciate, DALLE will give you more subtle and empathetic character expressions.
Midjourney, in comparison, wants to make characters look cool or beautiful, which means it tends to produce blank, fashion-model-like expressions or very on-the-nose emotions for a given scene (big smiles if it's happy, screaming terror if it's scary).
However, while DALLE is great for cartoonish images with quirky characters, it's not great for anything more atmospheric or emotionally complex.
So we'll usually do a second pass in Midjourney, which requires much more experimentation and produces wildly divergent results, much like coaxing a frustrating eerie genius cat.
Unlike DALLE, where you issue simple instructions and only tune the request when you bump into problems with the underlying world model, prompting Midjourney has become much more about mixing and remixing different images, including feeding images from DALLE and then getting Midjourney to apply its own art styles.
The process produces plenty of garbage, but then occasionally some startlingly beautiful images (so long as you don't look too closely at the details).
Here's a quick sample of alternating DALLE and Midjourney images so you can see the difference.
You get the idea!
And if you live in an interesting location or with a student population who are underrepresented in the image collection, feel free to get in touch and we'll make some more images to fit your students' brief!