Crafting AI Stories

Catching up with the latest AI news, I got inspired to try out some new concepts. So in this first example I’m playing with a half-finished project concerning a RPG adventure. It has some text and perhaps one day an AI will complete it for me. Until then, lets see what we can do. First attempt is a text prompt that got completed through GPT-3 (davinci-002):

The pond is said to be the home of a water spirit, who is said to grant wishes to those who are pure of heart. The spirit is said to be very shy and will only reveal itself to those who are truly worthy.

Next up is generating some art from stable diffusion. No problem – but it will be required to keep some kind of more specific style if used as part of a coherent story. So how about a kids book. Well let’s see how it goes:

All well, ok perhaps one more then, how to continue this? Last input into GPT-3 again:

The ledge is not large, but it is enough. Enough for a person to sit on and to be at peace with the world. The sun sets in the west, and the light dances through the trees and across the land.

Not entirely bad. It certainly capture some aspect of how I might have imagined the story. A bit of glue missing in between plus some colour correction. With some training and a lot of patience it might work out. But we have a lot to do in HCI if these tools are going to be how creative people work from now on. There are a lot of things that needs to be studied.