Follow along, if you like!
I’m using a walkthrough, because I can’t remember how to solve all the puzzles 42 years on, not that I ever DID solve all the puzzles back in the day anyway!
I was all of 8 or 10 years old when I first played Zork, I think, in 1982 or ’84, on a PC at a friend’s house that I didn’t have infinite time on, so give me a break. I also had a Choose Your Own Adventure-like book adaptation of the game, which I “played” through many times when I was the same age. It wasn’t published by the CYOA people, but it was the same concept — Choose Your Own Adventure stories are basically a print form of the text adventure computer game in many ways, although a bit more limited in the choices the player can “input”. Zork was one of the most popular home computer games in the early 80s, at a time when tabletop role-playing games like Advanced Dungeons & Dragons were at one of their early peaks in popularity. Together these games created the cornerstones of the geek subculture, a movement which has blossomed and thrived since then, particularly as the internet took off in the 1990s.
DALL-E is a text-to-image AI developed by OpenAI, that uses natural language inputs to generate high quality images. It’s been growing in popularity in recent weeks as the internet has begun to discover and share the images created by it. A twitter account @weirddalle is worth a follow, if you like that sort of thing. (Which, who possibly couldn’t?)
I’ll be feeding the text of Zork through DALL-E as input, and the results, will be the images that I tweet along the text from the walkthrough on the twitter account for the project.
I get rate limited to 50 posts every 23.5 hours on DALL-E, so each time I hit my limit, I’ll have to take a break. Accordingly, it’s expected that this will take maybe a few days or weeks to complete. It’s also possible that I could run afoul of DALL-E’s anti-abuse filters with some parts of the game, and if that happens I will failover to CrAIyon, the DALL-E Mini AI. It doesn’t generate as good images, but it’ll do as a backup.
I’m really pleased with this project, it’s so simple and the execution is easy, but it’s fun, and I feel like a creative guy just for having the idea to do it. Simple ideas really make me happy.
To put Zork, one of the earliest PC text adventures, which was released some 42 years ago, into an AI-based text to image generating system, and see what it outputs for illustrations seems like the funnest, coolest thing you could do, and a great way to tie the cutting edge of technology to some of its early roots.
Not all of the images DALL-E will generate will be accurate to the game, and that’s OK. It’s fun just to see what it comes up with, using the sparse descriptions that the game gives. Most of Zork took place in your imagination, and so we get to see what an AI might imagine.
The downfall of this process is that DALL-E will not remember from one run to the next all the context from the previous events in the game, so it will in many cases forget things that it should be aware of, resulting in some odd continuity. But that’s not the point, of course. The point is to do something fun with technology, playing with it to see what happens.
If you want to play Zork for yourself, you can do that! It’s free to play in your browser through an embedded DOSBox emulator.