| Mobile Web

Google\'s Lyria 3 music-generating AI arranges well but still sounds awkward in test

Google’s music-generating AI model Lyria 3 can produce about three minutes of audio with vocals and lyrics from text or images, with structure set by prompts. In a test, it generated a track reminiscent of Kendrick Lamar’s HUMBLE despite changes to the key, and handled a genre shift to orchestral accompaniment smoothly. But lyrics often did not fit melodies, and a jazz-style test sounded like a beginner’s improvisation, with repeated emphasis on a seventh tone. The reviewer said the music lacked a clear motif and narrative shape.