Uploaded October 2020 | Updated November 2026, 3 hours ago
I trained OpenAI's "Jukebox" model on a 2-second clip of speech-to-text audio counting to 5, and I asked for 60 seconds of pop music based on that clip.This video contains 8 attempts.
Some highlights:
0:37 - demon voice
0:45 - the gates open leaking darkness from the void
2:24 - a haunting ballad plays softly in the distance
2:30 - drum fill
3:27 - power pop chorus
3:50 - guitar solo
5:12 - mirror universe dream pop bossa nova
5:59 - recorder solo
6:19 - "HI!"
6:42 - demon voice returns
7:39 - applause
7:40 - shouty folk rock in a crowded room
I trained OpenAI's "Jukebox" model on a 2-second clip of speech-to-text audio counting to 5, and I asked for 60 seconds of pop music based on that clip.This video contains 8 attempts.
Some highlights:
0:37 - demon voice
0:45 - the gates open leaking darkness from the void
2:24 - a haunting ballad plays softly in the distance
2:30 - drum fill
3:27 - power pop chorus
3:50 - guitar solo
5:12 - mirror universe dream pop bossa nova
5:59 - recorder solo
6:19 - "HI!"
6:42 - demon voice returns
7:39 - applause
7:40 - shouty folk rock in a crowded room










