Inspiration
We were inspired by how much fun AI Dungeon was, so we wanted to create a journey where users could visually see how their decisions were affecting the story. We also wanted to solve the problem of commercial visual novels and stories that end too quickly and leave the reader unsatisfied.
What it does
AiVN is a visual novel game that is AI generated using the vivid imagination of its players. Visual Novels are digital and semi-interactive fiction games that focus on text-based narration and mostly static illustrations. AiVN generates a visual novel that can continue for as long as is humanly possible.
How we built it
Using express.js, we created a server to communicate with a machine learning model for the storyline and pictures. We parsed all the information received and displayed all of it beautifully in simple HTML.
Challenges we ran into
Our initial obstacle was that Chat GPT does not have an API. We considered using Davinci instead of Chat GPTI as it had an API. However we realized that Davinci did not consistently return what we wanted. Even though Chat GPT was better at returning what we wanted it still took a lot of fine tuning to be able to prompt chat GPT to act the way we wanted it to more consistently. Furthermore, finding a way to standardize Chat GPT’s responses was a great hurdle, and we decided to use description tags in brackets along with regular expressions to solve this issue.
Accomplishments that we're proud of
We’re proud that we learned so much at this hackathon and that we designed the website with only basic Javascript, HTML, and CSS. We are also proud of all the great discussions and collaboration among us.
What we learned
We learned the fundamentals of API GET requests and fetching responses using express. We also learned how to use web APIs to effectively prompt machine learning models such as ChatGPT and DALL-E, in order to generate text and images. We practiced a lot with regular expressions for string parsing, and gained more experience with the fundamentals of Javascript, HTML, and CSS. Finally, we learned how to work effectively in a small team, distributing the work appropriately and using pair programming to solve common issues.
What's next for AiVN
Next we want to add save point bookmarks to return back to the same story by remembering the prompt given to the AI and the previous dialogue, AI voice acting, AI music, and autoplay. We also would like to find a way to minimize the time it takes for chat GPT to load the prompt or use another AI that would be faster but still consistent.
Log in or sign up for Devpost to join the conversation.