I haven’t felt this way since I launched my first web site 30 years ago (May 27, 1995).
Or maybe when I was 12 years old, coding my first digital game (Programmer’s Revenge, a pseudo-chat bot that allowed the player to curse endlessly at their computer).
This morning I launched a web site I did not build featuring code I did not code for a game I designed but whose visual assets I did not created. It was all done in collaboration with AIs.
TLDR
In only ten days I vibecoded and posted a digital game using AI-generated videos inspired by the AI-generated cards in my tabletop game Uncannny Valley as a proof of concept. Please play Uncannny Valley here and let me know what you think.
THE BACKSTORY
In November of 2023 I started playing around with Midjourney, an image generating tool. It was just curiosity at first. But soon, it became an obsession. Since then, I’ve generated close to 8,000 images. Most of them weren’t for fun or art’s sake—they were for a card game I was designing, called Uncannny Valley.
Uncannny Valley is a tabletop game inspired by two modern classics: Dixit and Spot It. If you’ve played either, you know how different they are—and that contrast sparked an idea.
Dixit is all about surreal storytelling. Each card presents a strange, dreamlike scene with no direct relationship to the others. It’s a poetic exercise in interpretation.
Spot It, by contrast, is pure math in disguise. Each card shows a random-looking collection of symbols—scissors, ink blots, ladybugs. But here’s the trick: every single card shares exactly one item with every other card in the deck. Just one. Always.
I started wondering: what if I could blend the surreal world of Dixit with the precise math of Spot It? Could I create scenes—actual illustrations—that looked like spontaneous AI-generated art but were, in fact, carefully constructed to contain one and only one overlapping element with every other card?
To do that I had to unravel the magic behind Spot It.
I found answer, it turned out, in deep, beautiful math.
COMBINATORICS & AI
When I researched the science behind Spot It I fell into the rabbit hole of combinatorics. Combinatorics is an area of mathematics primarily concerned with counting. In the image below, on the right, we see how to visually represent 7 cards, each containing 3 objects, in such a way that every set of three shares one item and no more than one item with every other set.
Look at the apple at the top. It is grouped with RED LINE (apple, lemon, pair), BLUE LINE (apple, orange, cherry) and GREEN LINE (apple, banana, lime). And while apple is NOT grouped with the other 4 lines (YELLOW, LIGHT BLUE, ORANGE, & PURPLE) the apple’s other two partners are.
I couldn’t understand HOW it worked but I could see that it did. With 7 cards each containing 3 items. But how to design 25 cards each containing 6 items?
Turns out, I couldn’t do that in my head. So I turned to spreadsheets.
Each card’s “scene” needed a defined set of objects. I used combinatoric logic to determine which elements would appear on which cards. And I used a spreadsheet coded to take a list of items with a set number of categories and populate them–combinatoric style–each in their own row. And the end of each row combined those elements into a sentence to be entered as a prompt into an AI. For example: “A top hat and a little boy as dancing ballerinas, in the mountains, in the palette of yellow, as watercolor”.
Once I had the data, it was time to generate the art. That’s where Midjourney came in. I’d take the visual prompts and run them through Midjourney again and again, tweaking the results until I found one that felt just right. Sometimes it would take hours to get one image strong enough to represent all of the element. And my rule was it had to work on the own, with no editing from me. Each card would be a record of how the AI, at the time, interpreted the prompt.
FROM DECK TO DOZENS OF GAMES
Over time, I’ve created four complete themed decks, and dozen of ways to play with them. Each one got closer to my goal—not perfection, but something more interesting: cards that reveal the difference between how humans and AIs see the world.
That’s the heart of Uncanny Valley: it’s not just about playing a game. It’s about developing visual literacy. Can you sense when something feels just a little… off? Can you learn to describe the uncanny valley where the differences between how we and the AIs understand the world become difficult to ignore?
I posted cards on Instagram and invited people to sign up to receive a copy for playtesting. I used the deck for the past few years with educators to teach how game can be used to teach A.I. literacy.
And I thought that was the end of the story. It turned out that was only the end of the first chapter.
VIBECODING VIDEOS
Ten days ago I learned that Midjourney added a feature which creates short movies. Even better, I could set the image for the first frame. That meant all of the cards I had crafted for Uncannny Valley, all of those static images, could now “come to life.” Many of the cards captured a moment in time, which I suspect is why I found so many of those images compelling, leaving me wondering, What is going on here?
Now, with the videos, I could find out.
It did not take me long to ask how a version of the game that was digital and animated would change the learning and play experience. To find out, I turned for the first time to ChatGPT’s coding function.
I knew it was there but had no reason to check it out. I told ChatGPT it was an intern and I was giving it an assignment to build a new game and to teach me how to launch it on the web:
You are an intern at a game design company. I give you a task. I want to adapt a board game to a web-based game. In this game you are given four cards. Each card has a scene on it. Each scene is composed from the same six categories. There are five options within each category. When cards are put side by side, due to Combinatorics, there is always one and only one item that matches between the two cards. In the game, the player puts out four random cards and the first player to find and identify a match wins the round, gaining one point. Game plays until someone has 11 points. For the mobile version of the game, the physical cards will be replaced with videos which each contain the same properties of the cards. Design a prototype of the game for me that lets me put videos into a folder, the game will randomly pick four videos from that folder and display them in a 4 X 4 grid, and provide a playful and accessible interface for the player to first identify the two cards and then quickly name the element they have in common, give them a point if they are right, lose them a point if wrong, and then give them a chance if wrong to guess again but if they are right to replace the two selected cards with two new randomly selected videos from the folder. Attached is a list of the categories and the options within each one. The game is called Uncannny Valley. Have any questions?
After a few false paths, without me understanding 90% of how the code functions, I launched the site this morning. Don’t get me wrong–ChatGPT made it possible but it was still a lot of work: figuring out how to get the AI to give me what I wanted, in a format I could work with, with directions I could follow. For almost two days at one point the game was broken, and I lacked the coding knowledge required to fix it; then I realized I could ask ChatGPT to teach me how, and it did. But as I said – like coding my first web site in 1995, like programming my first games as a boy, this process delivered the same level of dopamine excitement found in creating and solving my own challenges.
This time, however, I wasn’t solving them on my own. ChatGPT congratulated me this morning once I told it the site was up and working. Note the last line, in which it makes a punny joke and does a call-back to its prior knowledge of my seltzer expertise.
So please go check out Uncannny Valley and let me know what you think (and play the video below).
