Infinite Piece

• 

  First scene of the game

• 

  Another AI generated scene

• 

  A generated scene

• 

  Another AI generated scene

Inspiration

Visual novels using have very limiting choice. We wanted to make one where the story can take any possible turn and the only limit is the user's creativity.

What it does

You start with an initial setting, and are prompted to input your character's actions. Then, this input is used to generate the next scene in the story with ChatGPT, and an image to describe that scene using Stable Diffusion.

How we built it

To get the stories in the right format that we could parse in our game, we created a dataset and fine-tuned ChatGPT. Then, we created the game using Renpy first, but then moved to PyGame since we could not include our AI generated images otherwise.

Challenges we ran into

• Having ChatGPT be consistent wtih the format so that the game would not break. Having it create actually plausible stories.
• Generating images and displaying them with the dialogues on Renpy (which is not possible)

Accomplishments that we're proud of

We managed to make a tough choice and migrate our game into Pygame very late into the hackathon and still managed to get a working product. We can now play and create our own stories and have fun with it.

What we learned

We learned how to fine tune ChatGPT with our own dataset, how to use Renpy, how to use Pygame and how to use Leonardo AI's API to generate images with Stable Diffusion.

What's next for Infinite Piece

A better image generation model, since the images are not as good as we wanted them to be. Also, ChatGPT is not very consistent with the quality of its stories and dialogues. Sometimes they are great and sometimes very bad. It also doesn't always respect the format we set for it.

Built With

• ai
• chatgpt
• diffusion
• pygame
• renpy
• stable