Google DeepMind

Lyria 3.5

Google DeepMind's full-length music model — describe a song and get back a 3-minute track with real vocals and arrangements.

Browse Lyria 3.5 prompts →Compare side-by-side

0 benchmarks run

Maker

Google DeepMind

What is Lyria 3.5?

Lyria 3.5 is Google DeepMind's most capable music generation model, turning a text prompt into a complete stereo track with sung lyrics, full instrumentation, and emotional vocals — up to three minutes long. It launched in July 2026 in Google Flow Music and expanded to the Gemini API in September 2026, with every output carrying an inaudible SynthID watermark. BPM control, duration matching, and exportable stems give producers meaningful hands-on creative control.

Strengths

✓Full-length songs up to 3 minutes with coherent structure — not just short clips.
✓Expressive, emotionally nuanced vocals with clear pronunciation and strong lyrics prompt-following.
✓Precise creative controls: BPM steering, exact-duration matching, and stem exports for production use.

Weaknesses

✗Text-to-music only — no audio-in continuation or remixing of an existing track.
✗Nuanced genre blending and fine stylistic control can be inconsistent.
✗Tied to Google's ecosystem (Flow Music, Gemini API) — no standalone open model weights.

Want to use Lyria 3.5 on your own photos?

We host Lyria 3.5 alongside every other major AI model. Free to try.

Open the Lyria 3.5 prompt library →