Billie Eilish’s “Bad Guy” just passed one billion views on YouTube. But even more impressive than that are the tens and thousands of fans who’ve uploaded their own versions of the song.  These covers span nearly every genre, language, and country. They range from professional to amateur and from ridiculous to sublime; revealing the best of the YouTube creator community’s breadth and diversity. Infinite Bad Guy is an AI Experiment that uses machine learning to bring thousands of these covers together, seamlessly aligning them in endless  combinations, creating a music video that’s unique with every view — in other words, an infinite music video.

Infinite Bad Guy uses audio alignment techniques to seamlessly stitch together thousands of fan covers into one explorable, ever-changing music video. We trained a neural network to take any cover version of Bad Guy and recognize which section and beat it corresponds to in Billie’s original “Bad Guy.” Once we’ve identified those similar beats sections, we can align them with the original song, and with other covers, allowing users to switch seamlessly between versions. By using a custom-trained neural network, we can take covers that are played using entirely different instruments, tempos, or even played in different order, and align every beat of them with the original song.

All the cover videos are analyzed with Google’s Knowledge Graph, which identifies characteristics like instruments, genres, and more. This makes it possible to explore covers that feature saxophones, sound like heavy metal, or take place inside a video game.

The end result is a seamless, explorable musical experience with near-infinite combinations, so many that before you watched all of them, the universe would likely have ended.