Inspiration

The idea of combining different animal parts to create a monster came from fairytale creatures like unicorns and dragons as well as Frankenstein. The inclusion of white outline was inspired by the way ransom notes in media use cut out magazine pictures.

What it does

We have a logic file that handles routing information between the important frankenstein.py and parts_classification.py. It first gets the input from the user and then hands that off to parts_classification.py. There we use Gemini to turn the user request into a standardized json like format that is matches phrases to animal parts. Then logic.py translates this into different file names corresponding to those parts and hands that off to frankenstein.py. frankenstein.py then grabs the relevant files and builds the resulting animal layer by layer using some vector addition we calculating using anchor points and socket joints stored in a large .json file. Finally we output that png to the user in the website.

How we built it

After settling on our idea we chose 10 different animals to use that we cut up into pieces throughout the project as we needed them. We then figured out how to construct animals from a composite of different parts with vector addition. From here we split off into working on the handling of user input and the construction of the creature. For construction we created a skeleton to create animals using random parts available and slowly introduced more parts are more slices got finished, and for the user handling we worked on creating a format that could be easily parsed into any information we might need. Once both were finished we worked on the logic function that would be taking in the user input to route the information through the different files until it got the final result.

Challenges we ran into

Separating the animal pngs into their respective parts was very time consuming
Anchor points needed to be manually fine tuned in order to prevent different parts from being egregiously disconnected from their intended socket
Fine-tuning the LLM's user input, so that it maps the correct body parts to animals

Accomplishments that we're proud of

Getting a mostly functional product done in time

What we learned

Aligning all the animal into one animal that make a unique trend across the young adults to print in their outfits and addon illustrative artistic space.