Imaginate-AI

Imaginate-AI - An educational game built on text-to-image generation

Inspiration

Our inspiration for this project was the recent public release of OpenAI's dalle-3 text to image model. After seeing how impressive the results were, we wanted to incorporate a text to image AI model into our own web application.

What it does

Imaginate is a game similar to the style of kahoot or skribbleio where users guess the caption used to generate an AI image.

How we built it

We built this web application by utilizing pretrained stable diffusion and transformer ai models from huggingface. We displayed the results of these AI models through a react frontend and a flask backend.

Challenges we ran into

The one major challenge we ran into is prompting the stable diffusion algorithm correctly to produce the best results. Text to image models are heavily controlled by the user defined caption so carefully constructing the captions with the correct number of inference steps and guidance scale is key to product high quality images.

Accomplishments that we're proud of

We are proud that our original idea from the start of the hackathon worked and is fully functional. We even spent a great deal of time playing our game because we truly find it fun and are passionate about it.

What we learned

The main thing we learned from this hackathon is how to seamlessly integrate flask and react to utilize python code and algorithms efficiently and effectively.

What's next for Imaginate-AI

There are several next steps we can take to improve Imaginate-AI. The first step would be to integrate multiplayer functionality, as of right now a user has to hide his caption from another player before handing over the game, and incorporating a multiplayer aspect to the game would fix this problem. Additionaly, we would expand the data we keep track of in our backend. We had ideas of storing a users games won, average score, total games played, and total captions guessed.