Hacker News new | ask | show | jobs
by riiii 517 days ago
One thing I'd like to see AI used for: draw the Zork scenery based on the in-game text description.
4 comments

Playing Zork with AI-generated imagery [video] (youtube.com)

133 points by mr_walsh on Jan 8, 2023 | hide | past | favorite | 54 comments

https://news.ycombinator.com/item?id=34300765

https://www.youtube.com/watch?v=ZpCrBBj6AWE

I think people have been experimenting with integrating stable diffusion into text adventure games (like AI Dungeon) for a while now.

Prompt adherence is always an issue particularly in a text adventure game where position matters - e.g. (The crystalline amulet is sitting on top of the old wooden table. A fireplace at the end of the room gently crackles. To the east lies a door, cracked with age. On the west wall you can see a painting of the lords of the manor. etc)

Flux could probably get close - though it clearly doesn't have any examples of an amulet in its training data. :)

https://imgur.com/a/giBqQ1L

That's a fun idea. The lack of continuity will be amusing, especially given the sparse descriptions of everything.
The later Infocom games — I’m thinking especially Moriarty’s evocative Wishbringer, Trinity, Beyond Zork — have much improved writing in general; it might be interesting to see what an LLM actually does with those.
A Mind Forever Voyaging has very good writing as well (and less in the way of puzzles). Of course, Hitchhiker's has great writing too.Not sure I ever played Wishbringer or Beyond Zork but Trinity is good.
That sounds like the perfect use for LLM image generation!
Somebody did this recently for Advent of Code (some year, not 2024 necessarily). I don't remember any details, but the images were entertaining.