What

Spot the Synth is a straightforward flashcard-style app, designed to practice spotting images generated by AI. My dataset is one that is commonly used to train image classification models to detect AI images; I figure if it's good enough to train a detection model, it should be good enough to train humans.

Why

My goals for this app are twofold:

AI-generated images are constantly getting better, and whenever a new advancement is made, the images look very realistic. But we get used to it. Think about special effects from the early 2000s vs now - people were genuinely blown away by the realism, but today we can all see that the CGI is stilted and outdated. But it takes repeated exposure, and I built this app as a sort of training routine.
I want to have a simple, but real-world, application deployed for people to make PRs against. I am a self-taught dev, and having a couple of collaborative projects on my resume helped quite a bit. Yes there are many open-source projects available for this purpose, but this project is intended more for people within my extended circles, for comfort and so that I can easily mentor and guide.

Features

A frontend that pulls a random image ID from a supabase table and uses Cloudflare Workers to display the image to the user, and then records information about their guess.

Planned features if I ever have the time:

Redis layer
Graphs to visualize guesses
Compete against a detection model
User profiles to save progress

Tech

Cloudflare pages
CLoudflare Workers (migrated from AWS Lambda; you can read about the process here)
AWS Lambda (original version)
Vite
Supabase