Dungeon:Here

Landing/Splash Page
Category Select Page
Character Name Page
Loading Page

Inspiration

The inspiration for Dungeon:Here stemmed from the team's passion for video games. During the Covid-19 lockdown, Dungeons & Dragon (D&D) was an outlet for our team to spend time together. It acted as a hobby, a creative outlet, and something to look towards to every Saturday evening. But with school starting again and people's schedule filling up, D&D simply took too much time and effort to run. That's why when we saw Co:here's NLP API as a challenge, we thought if we could make D&D fun and easy, then everyone could experience their own D&D adventure.

What it does

Dungeon:Here is text based adventure game that utilizes natural language processing to write an unqiue adventure each time based on user input. This allows users to create unlimited possibilities and stories each time they play. Dungeon:Here currently supports six genres to base your adventure on thats been trained on only 12 prompts we given it. Dungeon:Here will automatically generate a random starting prompt then users can decide what they want to do next. Dungeon:Here will then take user input and continue the story. Immerse yourself into the plot and spark your imagination.

How we built it

Dungeon:Here is built with Python, Html5, and Javascript. We used Python as our backend language with Co:here's Generate API as our NLP. Our frontend utilizes Html5, and Javascript to create our user interface.

Challenges we ran into

This project was a learning experience. It was our first time using NLP, using Stable Diffusion, even making a game. We initally ran into problems figuring out how to generate a proper reply after user input. With no time to train our own models, using Co:here's existing models made fine tuning the text generation to fit each genre a difficult task. Another challenge was to then pick out certain words that would be used in stable diffusion to generate the proper image. We finally got image generation but we just could not get a coherant image. Stable diffusion is a very interesting and powerful tool that we will continue to dive into and to fine tune the modifiers to create better images for the setting.

Accomplishments that we're proud of

This was the first hackathon for some of the team members and we are very happy we finished. Learning Co:here and Stable Diffusion has been a huge accomplishment for the team. We are happy that within the given amount of time we finished and the text generation is within the ballpark of what we were aiming for. The image generation isn't working as intended right now and it might not be perfect yet but we see a lot of potential in refining it more.

What we learned

We learned how to use NLP and Stable Diffusion to generate outcomes based on user input.

What's next for Dungeon:Here

Dungeon:Here needs a lot more work. We would want to add multiplayer, add a inventory system, and add more genres and characthers. Making the game more unique and have more features would make it more fun and allow more creativity for the users. However, the biggest improvements going forward are with the AI generation. We need to fine tune the text generation by training the AI on certain models for each genre. Create rule and datasets to feed into more prompt generation. It would be intersting to see if we can find another way to make the story more cohesive yet unexpected if we gave it more data and prompts to work with. We would really want to get Stable Diffusion to work on proper AI art generation and speed up the process right now. We managed to get images but they just dont make any sense right now and that has to with modifier selection that we have to fix.