Hi! I am building a user-generated open world where any user can play with an NSFW AI character or add to the lore of the place to a never ending story and game. We are live at GetIdol.com. We are using NSFW stable diffusion and Llama models.
My interest in particular is how we manage AI workflows to create interesting game mechanics. Would love to share some tips we found out. As background, me and my team used to work on AI models for something much more boring (credit analysis). We built our own CNN to do so.
One of the interesting aspect of game mechanic design for me is how we move away from static deterministic models to more qualitative and human outcomes using AI. For example, we belong to the eroge or visual novel school of games. These games are normally navigated via a pick your own adventure style journey (choose response A, go to B…etc.). We wanted something more open world so some mechanics we are piloting now are sentiment analysis (e.g. does the Idol love you? how much does the idol love you? how consistent is its feeling?). We use this to input a trigger condition for clearing a level.
Another thing that I quite enjoy is building scenarios and then stringing together scenarios for a chapter. For this, we are using our image models and keeping the seed constant so that images generated are closer to a scene (avoid teleportation from one place to another in the imagery).
We are trying to build our game mechanics around subjective outcomes using AI. Super excited about this area and love to connect with others trying something similar!