I think it'd be neat to take an old text only terminal game, and just auto generate the pictures based off responses.
Honestly, I probably wouldn't, I don't want my mental vagaries of what a Grue looks like to be set by some random image.
An interesting thing about CLIP is that when it doesn't know what something looks like, it instead generates pictures with the search text in them. That's why it confuses "an iPod" with "a piece of paper with iPod written on it".
The results were quite something - https://m.imgur.com/tfWLsSR
.club was on sale for $1.17