New Game Demo

My goal with this game, relative to the other games I've released,  is to explore how possession could work with AI powered narration. Specifically, what a game would look like where you may not be fully in control all the time. To this end, I've built out a global Gensokyo sandbox that progresses for all characters  without any need for player input, albeit without narrative depth (each character engages in one of a few dozen activities, several types of events occur). This global simulation then ties seamlessly into the local module of the game, representing the story unfolding from the player's perspective, with the LLM fleshing out the broad strokes of the global model into detailed narration.

Initially, you start as a passenger only able to slightly influence your host as they go through their day, but, as you develop new abilities, you can explore a variety of possession archetypes, ranging from becoming a voice in your host's head to controlling them outright. 

The ideas described above are not fully realized in the demo--the relationship between the global and local model, for example, can be a bit tenuous. That said, I'm happy with the progress I've made.

Another departure from previous games here is that this game is explicitly based on an existing setting--Touhou. I want to be frank here, as I think Touhou fans may play this game and find it wanting: my choice to use Touhou was pragmatic, and I don't have a deep background with the Fandom (though I know much more about it now than I did at the start.)

To realize my goal for this game, I needed characters with huge libraries of illustrations of them individually, and, critically, of each of them interacting with the others. I also needed a deep corpus of lore to construct detailed biographies and relationship histories between characters so that the characters could credibly interact without depending on the player to drive the narrative. In practice, the image requirement was the tighter constraint. Analyzing the boorus, the only possibilities were Touhou, and, far behind it, a few of the big gachas. I'm not a big fan of gacha so my choices were to make a Touhou game or to drop the project. I proceeded despite reservations about whether I could do the setting justice. As such, I'd be grateful for input from Touhou fans to give the game more personality and heart and handle the setting more competently.

While this game could be called complete except for the missing endgame, many features are unfinished. I want to be more ambitious with the global simulation, making character logic more complex, more detailed, and more relevant to the local module. I want to make location differences more important. Many of the game mechanics are currently awkward and need to be reimplemented or redesigned. Dynamic portraits for both host and focus character are not yet implemented, and the image serving engine is unchanged from Orgone Collector--I plan to use a bunch of tricks to improve image serving when two characters are relevant to the image, but I have not implemented any of them yet. I also need to refine context handling to improve LLM adherence to gamestate and logic.

I'm excited to hear what people think of this experiment.