Lyria 3.5 brings full songs with vocals to the Gemini API
Two preview model IDs turn a prompt into a multi-minute track or a 30-second loop, in any language — every clip SynthID-marked, licensing still murky.
Copy markdown
Full songs, vocals and all
The lyria-3.5-pro-preview model turns a text prompt — plus up to 10 reference images — into a multi-minute track with real verses, choruses and bridges and expressive vocals. Output is 44.1 kHz stereo, and lyrics come back in whatever language you prompt in.
A 30-second loop model, too
Its sibling, lyria-3.5-clip-preview, returns fixed 30-second clips — the right primitive for game loops, UI stingers and ad beds you can drop straight into a build without trimming a full song.
Ship it this weekend
It's the same Gemini SDK you already call, so a text-to-music endpoint is a few lines. Section tags like [Verse] and [Chorus] plus timestamps give structured control — enough for karaoke timelines, dynamic game soundtracks, or auto-scored video intros.
Read the rights before you ship
It's preview-only, and a SynthID audio watermark is baked into every track. That covers Google's EU labeling duty, not yours — deployers carry separate Article 50 obligations, and the training-data question is still open after a Munich court ruled against Suno in July.