Spicy Ginger Labs
Built at Meta Global AI Developer Hackathon · Oct 26, 2026 · Remote

Popsmosis turns any topic into a "droplet": a fact-checked study song and a narrated lesson video, made automatically in about 30 minutes. A student types a question, like "How do the lungs get oxygen into the blood? For AP Biology", and can pick a song style, like K-pop, or add a photo of their notes. An AI teacher picks up the request within seconds and runs the studio, a pipeline I built during the hackathon. Meta's Muse models plan the lesson, write the text, narration and lyrics, then fact-check every line, so a catchy rhyme that would teach something false gets rewritten. Muse Image paints each scene. SAM 3.1 finds every part in it, so labels, spotlights and moving particles land in the right place. Meta's Demucs and MMS time the sung words for karaoke lyrics. My own renderer draws the videos frame by frame. The problem: SAT Sprints, the AP and SAT prep platform I built, teaches in text, and text doesn't click for every student. Lesson videos and songs help lessons stick, but AI-made ones get facts wrong, and students learn the mistakes. Making them with AI still took a lot of human intervention: wrong pictures, wrong lyrics, rounds of fixes. Popsmosis automates all of it, and it's honest about what it can't be sure of: every picture is marked AI-generated, and when a check finds a mistake, the video says what's true. Finished droplets go on a public Droplets tab, like YouTube for study videos, with hashtag search and a Clips feed for phones. Anyone can watch for free, and anyone with a Google account can ask for their own and pick the brain that writes it, Meta's Muse Spark included. Before the event: Popsmosis started from zoey, an existing AI advice-column app, for its sign-in and question queue. Everything else was built during the hackathon. Live: popsmosis.com · How it works: popsmosis.com/how-its-made