Google's Lyria 3.5 brings full 3-minute songs to Gemini and the API
Vocals, custom lyrics and a SynthID watermark ship to the Gemini API — as a Munich court rules Suno infringed and Deezer sees 90k AI tracks a day.
Copy markdown
Full songs, not 30-second clips
Lyria 3.5 generates cohesive tracks from a 60-second clip up to a full three minutes at 44.1kHz stereo — verses, choruses and bridges, in your choice of genre and vocal or instrumental. It's live for every Gemini user globally on web and mobile, a clear step up from the old clip-only model.
It's in the API — script music into your app
Two models hit the Gemini API: Lyria 3 Clip for 30-second loops and Lyria 3.5 for full songs. You can pass custom lyrics with structural section tags, up to 10 reference images, and instrument entry timestamps — enough to auto-generate backing tracks, brand jingles or personalized ringtones this weekend.
SynthID on every track, hard limits on voices
All output carries Google's imperceptible SynthID watermark, and the model refuses requests for specific artist voices or copyrighted lyrics. Generation is single-turn for now — no cross-prompt editing, so you re-roll rather than tweak.
The copyright backdrop builders should watch
The launch lands weeks after a July 31 Munich court found Suno liable for infringement, asserting jurisdiction over US-only training and rejecting a fair-use defense. Deezer now says ~90,000 fully AI tracks arrive daily — over half of all new uploads — which is exactly why Google is leaning on watermarks and guardrails.