-
The initial page
-
Uploading an image
-
A generated image and recipe for a squash slice
-
A generated image and recipe for a beanie
-
The history page for all recipes
Inspiration
I was browsing reddit and I saw an image of apple slice-shaped cookies. The idea then just popped in my head suddenly.
What it does
With an uploaded image, it changes the content of the image into a cookie, and provides a recipe to make that cookie. The history of every generated recipe is kept on the start of the server.
How we built it
I divided the parts into the frontend and backend. Some parts of the frontend were vibe-coded via GitHub Copilot, since I don't have much talent in UI design. I then wrote the image uploading part to ensure input worked. Next, I built the pipeline of how the uploaded image is run through models to the get the desired output. I used Gemini API for the models. I first used Nano Banana to convert the uploaded image into a cookie version, then regular Gemini to generate a cookie recipe based on the latter image. The recipe and image generated in the backend would then be sent to the frontend to be displayed.
Challenges we ran into
The primary challenge was getting the model calls set up and making sure they worked, since I didn't know how long it would take to generate something. Figuring out the correct prompts to make the desired image also took a while. The other issues were for the frontend, since the vibe-coded code needed several edits as development went on.
Accomplishments that we're proud of
I'm happy that I managed to finish a prototype of the project.
What we learned
I did learn a lot on using the Gemini API, as well as the limitations of vibe coding when it comes to describing what you want. It also costs about 2-4 cents per recipe generated, so the web app is not very cost-effective.
What's next for ImageToCookie
Improve the image detection capabilities and find a way to reduce token usage.
Log in or sign up for Devpost to join the conversation.