Google’s generative AI music platform using Lyria 3 for songs and Veo for music videos, with a vibe-code layer for building custom audio tools.
Key Takeaways
Lyria 3 handles full-length song creation; Veo powers user-directed music videos with controllable characters and aesthetics.
“Vibe-code” lets you build audio plugins, music games, and custom DAWs via prompts, no traditional dev environment needed.
Stem splitting, audio effects, and daily credits are included; personalization model trains on your output over time.
Free to start with no credit card; publishing, playlists, and artist-following are built in.
Hacker News Comment Review
Strong consensus that song generation defaults to generic EDM or dubstep regardless of prompt; commenters trying ambient, lofi, or R&B got similar results, suggesting weak genre steering in Lyria 3.
The vibe-code/instrument building path drew more genuine excitement than song generation – one commenter built a working wave-blending drone synth casually in a single session, pointing to real utility for builders and sound designers.
Non-western and microtonal music is unsupported in practice; commenters attribute this to western-biased training data, making genres reliant on microtonality (e.g. maqam, Indian classical) unreachable.
Notable Comments
@dgalati: Built a wave-blending synth on the couch via prompt; calls it “really sweet as a tool for creating ways to create music” – distinct from song generation quality.
@inerte: Tested microtone-dependent genres; model ignores them entirely, producing Polyphia-style western output instead – frames non-western music as a structural blind spot, not a UX issue.
