Lyria 3.5 opens to devs: full songs with vocals via the Gemini API
The Gemini API can now score your app — multi-minute tracks, timed lyrics, section control, from text or images. Plus Meta's Muse agent goes mainstream.
Copy markdown
Vocals, timed lyrics, and structure from a prompt
`lyria-3.5` generates multi-minute songs at 44.1kHz stereo with real vocals, timed lyrics, and full instrumental arrangements — steer the structure with `[Verse]`/`[Chorus]` tags and timestamps. It takes text or up to 10 images, works across languages, and exports MP3 or WAV.
Three ways in — and a Clip-vs-full-song split
Reach it from Google AI Studio, the Gemini API, or the Gemini app. Two variants divide the work: `lyria-3-clip-preview` for fixed 30-second loops, and `lyria-3.5` for full multi-minute tracks with distinct sections.
Build this weekend: a soundtrack that hits its cues
With lyric, section, and timestamp control you can generate a game or video soundtrack that lands the chorus exactly on a beat drop, or turn a brand image into a jingle — one API call, no DAW or session musicians.
The catch: one shot, and it's watermarked
It's single-turn — no iterative editing, and the same prompt yields a different take each run. Every output carries a mandatory SynthID watermark, and artist mimicry plus copyrighted lyrics are blocked.
Elsewhere: Meta's Muse agent goes mainstream — and pays
Meta launched Muse broadly across web, mobile, WhatsApp, and glasses at $20/mo (Power) and $100/mo (Maximum), on Muse Spark 1.3 in a Secure VM. The builder angle: it routes real transactions through Stripe and Shopify, pushing agentic payments to consumers.