From a Music Conservatory to the AI Studio
Years of classical vocal training leave a permanent mark on your ears. When Google rolled out Lyria 3 — their latest generative music model — inside Gemini with access for all 750 million monthly users, the news landed differently for those of us who spent years studying Baroque counterpoint and contemporary composition.
My first instinct wasn't wonder. It was "let's hear what you've got."
This review comes from that place: trained ears, real test scenarios, and 10 professional Eastern music prompts you can copy and use immediately.
What Is Lyria 3?
Lyria 3 is Google DeepMind's most advanced music generation model. Previously limited to YouTube Dream Track and Music AI Sandbox, it's now built directly into Gemini — no waitlist, no separate app.
Three key upgrades from previous versions:
- Auto-generated lyrics — Describe a scene and mood; Lyria 3 writes the words
- Better control — Genre, vocal timbre, tempo, and emotion can all be specified in your prompt
- Higher audio quality — Training expanded from 500K to 2M songs; output now at 24-bit audio, higher than most streaming platforms
How to Access It
Go to gemini.google.com → Tools → Music
- Available on desktop (mobile app rolling out)
- Must be 18+
- Supported prompt languages: English, German, Spanish, French, Japanese, Korean, and more — Chinese prompts not yet supported
- Free users can generate; paid tiers get higher limits
The 6-Element Prompt Formula
Before getting into test results, here's the prompt formula that consistently produces the best results:
| Element | What to Write |
|---|---|
| Genre & Era | The more specific, the better — "80s synth-pop," "lo-fi jazz" |
| Mood & Theme | What emotion or story? "Longing for childhood" works better than "sad" |
| Vocal Style | Raspy baritone, breathy soprano, no vocals at all |
| Tempo & Rhythm | BPM number + feel, like "82 BPM, slow swing" |
| Key Instruments | Erhu, guzheng, piano, vintage drum machine |
| Scene Detail | One specific memory or setting — the more real, the better |
One-sentence formula: Scene + Mood + Genre/Era + Vocals + Tempo + Instruments
Real-World Tests: Three Scenarios
Test 1: Late-Night Subway Commute BGM
The goal: a song that captures the feeling of an exhausted commuter on the last train home — not inspirational, not depressing, just that quiet "another day, let's keep going" energy.
Prompt used:
A Mandarin pop ballad in the style of Eason Chan's mid-career introspective works — think "Shall We Talk" or "Good To Be Bad", produced by C. Y. Kong. Warm Rhodes electric piano playing sparse broken chords as the harmonic spine, the way Kong layers a muffled electronic pulse underneath acoustic instruments. Male vocals in Eason's register — mid-range baritone, slightly nasal, restrained, never belting, delivering lyrics as if talking to himself on a late-night subway. 82 BPM, slow swing feel. A single cello line enters in the second half as a counter-melody. Subtle ambient layer of distant subway announcements and train-door chimes mixed very low. No drum kit — only a lo-fi drum machine with brushed texture. Melancholic, urban, intimate, like a conversation you have with yourself at 11 PM on Beijing Line 10.
Why this reference: Eason Chan's collaboration with producer C.Y. Kong has a signature restraint — piano as skeleton, electronics as atmosphere, strings used sparingly. Never sentimental, but always leaves you silent afterward.
Verdict: The instrumental texture surprised me. The interweaving of ambient city noise and piano was genuinely atmospheric. But the lyrics fell flat — Lyria 3 understood "exhausted" but not "Beijing exhausted." It felt like a sympathy card from someone who's never taken that train.
Conclusion: Excellent as an instrumental layer. Don't expect it to write lyrics that cut deep.
Test 2: 2 AM Contemplation
The goal: music for sitting alone at 2 AM — thinking about everything and nothing simultaneously.
Prompt used:
An ambient piano piece in the style of Ryuichi Sakamoto's late works — specifically the aesthetic of "async" and "12". A single upright piano recorded with room microphone, capturing the natural reverb and the mechanical sound of keys and pedals. No quantization, no click track — the tempo breathes freely around 52 BPM, slowing and accelerating like human thought. Sparse, deliberate notes with long silences between phrases. A subtle layer of field recordings underneath: distant city traffic hum, a window rattling softly, the faint buzz of a refrigerator. No drums, no bass, no synthesizer pads. In the second half, a single sustained cello tone enters and holds for 8 bars. Female vocals appear only as wordless humming, barely audible, mixed as if coming from another room. Melancholic but not sad. Still.
Why this reference: Sakamoto's late-period philosophy: silence between notes matters as much as the notes themselves. "async" uses preserved mechanical noise and ambient field recordings to create textures that are the most honest representation of 2 AM.
Verdict: The most satisfying result of all three tests. The floating quality was genuinely captured. Vocals were barely-there, like someone humming next door. The cello entrance timing was right — not intrusive. If this were a student submission, I'd give it a B+. Harmonic progressions played it too safe, but as an emotional carrier, mission accomplished.
Test 3: A Song for Mom
The goal: capture the quiet tenderness of a mother who expresses love through remembered meals and short phone calls that say everything.
Prompt used:
A Mandarin folk ballad in the songwriting philosophy of Jonathan Lee (李宗盛) — the restraint of "New Old Song" and the raw simplicity of "Hill". A single nylon-string acoustic guitar fingerpicking a simple arpeggiated pattern as the only accompaniment for the first half. Male vocals in a weathered, spoken-singing style — not technically perfect, slightly breathless, the way Lee delivers lines as if talking to you across a kitchen table. 63 BPM, rubato feel. An erhu enters only in the final third, playing a single sustained melodic line that mirrors the vocal melody one octave higher. No drums, no bass guitar, no synthesizers. The mix should sound like a live recording in a small room — you can hear the guitarist's fingers sliding on strings. The song is about a mother's kitchen — the smell of braised pork, her humming while chopping vegetables, phone calls that are always too short but say everything.
Why this reference: Jonathan Lee's method: fewer instruments = heavier emotion. His most affecting songs are just one guitar throughout. The erhu enters last for less than a third of the song — the most important instrument always arrives last.
Verdict: The first 15 seconds were moving. When the erhu entered, something stirred. But in retrospect, I was moved by the timbre of erhu itself — not by Lyria 3's emotional intelligence. The costume isn't the performance. Lyria 3 is clever about sound color; emotional depth still stops at the surface.
That said — if you want to make a small gift for your mom, it's more than enough. She'll love it.
Honest Assessment
What Lyria 3 Gets Right
- Instrumental arrangement quality exceeds expectations — Texture and mix layering is cleaner than most amateur producer demos
- Significant audio quality upgrade — 24-bit spec sounds better than most streaming platforms; nothing harsh on the ears
- Genuinely accessible — Anyone can use it without music theory knowledge
- Image-to-music is practically useful — Upload a photo, describe a mood, get a personalized track
Where It Falls Short
- Lyrics are inconsistent — Chinese logic and rhyme handling noticeably weaker than English; occasionally sounds like machine translation
- Emotional depth stops at "surface correct" — It won't surprise you; lacks the unexpected moments that define great music
- Only 30 seconds — Too short to function as a complete piece; works best as a BGM fragment
- Chinese prompts not supported — Must write in English; a real barrier for many users
Lyria 3 vs. Suno vs. Udio
| Dimension | Gemini Lyria 3 | Suno / Udio |
|---|---|---|
| Entry barrier | Very low — built into Gemini | Low — separate registration required |
| Generation length | 30 seconds | Up to several minutes |
| Lyrics quality | Inconsistent; English stronger | More polished, more stylistically varied |
| Instrumental quality | Excellent, 24-bit | Excellent |
| IP protection | SynthID watermark + content filter | Relatively weaker |
| Ecosystem integration | Seamless with image, video, text | Standalone tool |
| Chinese support | English prompts only | Relatively better |
Google's main differentiator isn't audio quality or stylistic variety — it's the entry point. You don't need another app. Inside Gemini, you can write, generate music, create cover art, and share in one continuous workflow. For most users, that's the real value.
10 Professional Eastern Music Prompts — Copy and Use
These 10 prompts form a complete collection: Wan Xiang Gui Yi — Eastern Timescapes. Each is modeled on the creative signature of a legendary master, covering historical narrative, ethereal ambience, martial tension, and imperial ceremony. Copy any prompt directly into Lyria 3.
01. The Mandate of Heaven
Style reference: S.E.N.S. (神思者) — masters of epic historical grandeur
A majestic Mandarin neo-classical orchestral piece. 72 BPM. Opens with deep, resonant Tanggu (Chinese barrel drums). A lyrical solo Erhu leads with exquisite vibrato, supported by a lush 60-piece string orchestra. High-end digital reverb for a vast, panoramic soundscape. Awe-inspiring and timeless.Best for: Historical documentaries, cultural heritage videos, epic narratives
02. Gap of the Firmament
Style reference: Joe Hisaishi (久石让) — ethereal folk melodies fused with symphony
An ethereal Mandarin fusion track. 60 BPM. Crystalline Guzheng and Harp duet playing cascading arpeggios. A breathy Xiao flute carries a melancholic melody. Build to a climax with soaring wordless female soprano vocals (vocalise). Pure, pristine, and spiritually uplifting.Best for: Meditation, yoga, healing content, nature cinematography
03. Sands of Zen
Style reference: Kitaro (喜多郎) — New Age master of the Silk Road's mystique
A mystical ethnic-ambient track. 80 BPM. Features a haunting Duduk or Suona (played softly). Intricate hand drum percussion (Tablas/Udu). Vintage synthesizer pads create a shimmering desert horizon effect. Mysterious, artistic, and visually evocative.Best for: Travel vlogs, exploration documentaries, mystery-themed content
04. Still Water, Lone Boat
Style reference: Ryuichi Sakamoto (坂本龙一) — minimalist piano and Eastern philosophical space
A minimalist Mandarin avant-garde piece. 55 BPM. Single upright piano recorded with high room resonance, paired with sparse Pipa plucking. No percussion. Long silences between phrases. Ambient water droplets mixed low. Introspective and deeply calm.Best for: Late-night work, reading, contemplation, minimalist content
05. Blade Through Shadows
Style reference: Tan Dun (谭盾) — wuxia tension and experimental ethnic scoring
A dramatic Mandarin martial arts orchestral track. 100 BPM. Staccato violins playing rapid motifs. Sharp bamboo flute (Dizi) trills. Powerful Taiko drum accents. A low-pitched solo cello provides a dark, tense foundation. Intense, cinematic, and sharp.Best for: Action videos, martial arts themes, tense cinematic scoring
06. Apricot Rain
Style reference: Sojiro (宗次郎) — ocarina master of rustic natural beauty
A tender, melodic Mandarin folk piece. 68 BPM. Sweet Ocarina or Xun melody. Accompanied by warm acoustic guitar and a light string quartet. Subtle sound of gentle rain and distant oars in the background. Nostalgic, peaceful, and heartwarming.Best for: Rural life content, parenting videos, warm daily vlogs, nostalgia
07. Roar of the Nine Heavens
Style reference: Epic orchestral — Hans Zimmer meets Chinese tradition
An epic Mandarin fantasy orchestral anthem. 95 BPM. Heavy cinematic percussion meets powerful male choir (Oohs and Aahs). A soaring Suona plays a triumphant theme. Layered brass and Taiko drums create a massive "wall of sound." Heroic and breathtaking.Best for: Game cinematics, hero themes, epic trailers, powerful openings
08. The Breath of All Things
Style reference: Himekami (姬神) — electronic layered over East Asian folk melody
A rhythmic Mandarin New Age track. 110 BPM. Plucked strings (Guqin/Sanxian) layered over a modern electronic beat. Nature field recordings (birds, rustling leaves) integrated into the percussion. Bright, rhythmic, and full of life.Best for: Modern city content, tech-meets-tradition themes, energetic lifestyle
09. Elegy at the Pavilion
Style reference: Shigeru Umebayashi (梅林茂) — "In the Mood for Love" aesthetic
A melancholic Mandarin tango-influenced ballad. 60 BPM (3/4 time). Solemn solo violin and melancholic accordion. Sparse Guzheng glissandos provide an oriental texture. Lush, dark string pads. Sophisticated, cinematic, and heartbroken.Best for: Art films, emotional storytelling, nostalgic themes, quiet elegance
10. The Golden Era
Style reference: Tang Dynasty court music with modern production aesthetics
A vibrant Mandarin traditional court ensemble. 88 BPM. Features Pipa, Sheng, and multiple flutes playing in harmony. Bright bronze bells and chime percussion. A festive yet elegant atmosphere. Rich, ornate, and culturally grand.Best for: Traditional culture showcases, festival videos, cultural pride content
Advanced Tuning Tips
If your generation sounds too "clean" and lacks texture, append any of these to your prompt:
Analog saturation for a vintage 90s record feel— adds warm, aged record textureHumanize the timing, avoid robotic quantization— makes rhythm breathe more naturallyThe vocal should be the most prominent element in the mix— pushes vocals forward in the mix
Final Take
Lyria 3 is a very smart tool. It is not a creative artist with a soul.
It understands "exhaustion" but doesn't understand "Beijing exhaustion." It can produce "warmth" but not your mother's specific warmth. Emotional precision needs more time.
But if you want a dedicated commute BGM, a funny custom song for a friend, or background music with the right feel — it's already more capable than most people expect, and easier to use than you'd imagine.
For those of us with formal music training, Lyria 3 is what I'd call a "distinguished student." Technically correct, no major errors, but not yet the moment that makes you put down your pen and just listen.
Best use case: Treat it as an inspiration tool and creative toy, not a replacement for commercial music production. The more specific and personal your prompt, the better the result.
Access: gemini.google.com → Tools → Music. Must be 18+. Write prompts in English.