You typed "an epic cinematic song with drums," hit generate, and got three minutes of the same wall of sound — no build, no drop, no chorus you could point to. The problem usually is not the model; it is the prompt. Lyria 3 Pro can architect a full song with intros, verses, choruses, and a bridge, but only if you tell it how to arrange one. A vague genre-plus-mood line gets you a vague genre-plus-mood loop.
Most Lyria 3 prompt tips online stop at "add more adjectives." That is not the lever that matters. The real difference between a generic result and a directed one is (1) filling every slot of Google's official prompt formula on purpose, and (2) using the timestamp technique — the underused feature that scripts section changes down to the second. This guide breaks both down, with four copy-paste example prompts across genres and one full worked timestamp arrangement.
Everything here is based on Google's official prompting guide for Lyria 3 Pro and the DeepMind documentation — plus what we have learned running the model behind the Lyria 3 Pro generator on this site.
The Official Prompt Formula, Slot by Slot
Google publishes one core formula. Memorize it and you will never stare at a blank prompt box again:
[Genre & style] + [Mood] + [Instrumentation] + [Tempo & rhythm] + [Vocal style & language] + [Lyrics]Each slot is a control surface. Leaving one out does not break the prompt — the model just guesses, and its guesses are where "generic" comes from. Here is what to write in each, and why it moves the output:
| Slot | What to write | Why it matters | Example |
|---|---|---|---|
| Genre & style | Primary category plus an era or fusion reference | Sets the whole sonic palette; fusions ("X meets Y") avoid the default sound | "1980s synthwave with modern indie-pop polish" |
| Mood | The emotional intent, in feeling words | Steers dynamics, chord color, vocal delivery | "nostalgic, hopeful, slightly melancholic" |
| Instrumentation | Name the actual instruments you want up front | Stops the model defaulting to generic pads | "gated reverb drums, analog synth bass, bright arpeggios" |
| Tempo & rhythm | Descriptive pace, not just a BPM number | Descriptions travel better than raw numbers | "mid-tempo, driving four-on-the-floor pulse" |
| Vocal style & language | Gender, texture, delivery, and the language | Controls timbre and which of the 8 vocal languages is sung | "airy female lead vocal in English, layered harmonies" |
| Lyrics | A theme to write about, or exact words after Lyrics: | Decides whether the model writes lyrics or performs yours | "a song about driving home at 2am" |
The most common mistake is over-loading Genre and Mood while leaving Instrumentation and Vocal style empty — the two empty slots are exactly where the model falls back on its most average output. Name two or three specific instruments and one concrete vocal direction, every time.
Rule of thumb: one specific detail per slot beats five adjectives in one slot. "Warm nylon-string guitar" plus "breathy female vocal in French" directs the model far more than "beautiful, gorgeous, emotional, stunning, heartfelt music."
Controlling Vocals, Timbre, and Language
The vocal slot deserves its own attention, because it is where AI songs most often sound off. Lyria 3 Pro sings in 8 languages — English, German, Spanish, French, Hindi, Japanese, Korean, and Portuguese — so name the language explicitly if you want anything other than the default.
Beyond language, describe three things in the vocal slot:
- Voice type: "deep male baritone," "airy female soprano," "raspy tenor."
- Delivery: "whispered and intimate," "belted and powerful," "conversational rap flow."
- Texture: "close-mic'd lead with wide stacked harmonies on the chorus."
For duets, describe more than one singer and Lyria arranges them together — for example, "a smooth male vocalist in English trading lines with a breathy female vocalist in Portuguese." It becomes a call-and-response arrangement, not two mixed-down takes.
Three flags change the output more than any adjective can:
- Instrumental tracks — end the prompt with the word
Instrumental.and no vocals are generated. Essential for backing tracks, scoring, and loops. - Your exact lyrics — write
Lyrics:followed by your words, and the model performs them instead of writing its own. Keep line breaks; they map to phrasing. - Duets and multiple voices — describe each singer's voice and language, and Lyria distributes the parts.
The Timestamp Technique: Directing Structure by the Second
This is the feature that separates Lyria 3 Pro from every "genre + mood" generator, and the one almost every prompt guide skips. Because Pro writes full songs up to 3 minutes with real sections, you can script the arrangement with timestamp cues from [00:00] to [03:00]. The model treats each cue as a guaranteed section change instead of hoping a chorus appears on its own.
Use timestamps whenever you need a guaranteed bridge, a build that lands on a specific beat, or a soundtrack that matches known scene changes in a video. Here is a complete worked example — a synth-pop track — that you can paste and adapt:
An 80s-inspired synth-pop anthem, nostalgic and hopeful, with gated reverb drums, warm analog synth bass, and bright arpeggios. Mid-tempo, driving pulse. Airy female lead vocal in English with wide stacked harmonies on the chorus. A song about leaving a small town to chase something bigger.
[00:00] Soft, shimmering synth pad intro with a single held vocal note.
[00:12] Arpeggios and a light drum machine fade in and build.
[00:24] Verse one begins — restrained lead vocal over synth bass and hi-hats.
[00:50] Pre-chorus lifts; drums open up and harmonies enter underneath.
[01:04] First chorus — full gated drums, stacked vocal harmonies, soaring melody.
[01:30] Verse two, slightly fuller than verse one.
[01:54] Second chorus, bigger, with an added counter-melody synth line.
[02:18] Stripped-back bridge — vocal and pad only, tension building.
[02:36] Final chorus explodes back in with all layers and an ad-lib vocal.
[03:00] End on a sustained, reverb-washed synth chord.The method is in how that prompt is built: the first paragraph is the plain formula (genre, mood, instrumentation, tempo, vocal, theme); the timestamped block is the arrangement. The formula sets the sound, the timestamps set the shape. You do not need a cue every few seconds — five to eight well-chosen markers (intro, verse, chorus, bridge, final chorus, outro) are enough to lock the structure.
Low-friction check before you spend a full generation: run just the formula paragraph — no timestamps — as a 30-second clip on the base Lyria 3 model first. If the genre, instruments, and vocal come back right, promote the same paragraph to a Pro generation and add the timestamp skeleton. You can test both in the Lyria 3 Pro generator and burn far fewer full-song generations this way.
Four Copy-Paste Prompt Examples (Different Genres)
These follow the formula and are ready to paste. Swap the specifics for your own project.
1. Lo-fi hip-hop (instrumental, for study/background):
Warm lo-fi hip-hop, relaxed and slightly nostalgic, with dusty vinyl crackle, a mellow Rhodes electric piano, soft finger-snapped percussion, and a round upright bass. Slow, laid-back head-nod tempo with swung hats. Instrumental.2. Pop ballad with your own lyrics:
A modern piano pop ballad, tender and heartfelt, with intimate grand piano, subtle string swells, and brushed drums entering in the chorus. Slow, rubato tempo. Soft male tenor lead vocal in English, close-mic'd and emotional.
Lyrics:
I kept the light on by the door
Long after you stopped coming home
And I still hum the songs we wrote
Even when I'm on my own3. Bossa nova / R&B duet (bilingual):
Classic bossa nova meeting modern R&B, intimate and warm, with nylon-string guitar, warm electric piano, brushed drums, and light shakers. Slow, swaying tempo. A smooth male vocalist in English trading verses with a breathy female vocalist in Portuguese. A duet about a long-distance summer romance.4. Cinematic orchestral trailer (instrumental, image-guided):
Epic cinematic orchestral trailer music, tense building to triumphant, with low pulsing strings, staccato brass stabs, a huge taiko drum section, and a soaring string melody at the climax. Slow build accelerating to a driving finish. Instrumental.For that last one, remember Lyria 3 Pro accepts multimodal inputs — you can attach reference images (up to 10) or a PDF to steer the mood and story, though it does not accept audio uploads or voice cloning. A concept-art frame often nails a cinematic brief faster than words alone.
Why Precise Prompts Pay Off Here
Lyria 3 outputs high-fidelity 48 kHz stereo, so a well-directed track sounds finished rather than sketchy, and every output carries an inaudible SynthID watermark plus C2PA content credentials — provenance matters if you publish commercially.
There is also a cost angle. A Lyria 3 Pro request returns a full song in a single generation, so on a credit-based studio you spend credits per track. Getting the prompt right on the first try instead of the fifth directly saves credits. For the full picture of what the model can and cannot do, our complete Lyria 3 Pro guide covers access paths, pricing, and limitations.
FAQ
What is the official Lyria 3 Pro prompt formula?
[Genre & style] + [Mood] + [Instrumentation] + [Tempo & rhythm] + [Vocal style & language] + [Lyrics]. Fill every slot on purpose; empty slots are where generic output comes from.
How do I prompt Lyria 3 to make an instrumental track?
End your prompt with the word Instrumental. and no vocals are generated. Everything else in the formula still applies.
How do I make Lyria 3 sing my own lyrics?
Write Lyrics: followed by your exact words, keeping line breaks intact so they map to phrasing. Without that prefix, the model writes its own lyrics from the theme you describe.
How does the timestamp technique work in Lyria 3 Pro?
Add cues like [00:00], [01:04], and [03:00] describing what happens at each moment. Pro treats them as guaranteed section changes, so you get a real arrangement instead of one continuous loop. It works because Pro writes full songs up to 3 minutes with structure.
Why are my Lyria 3 prompts giving generic results? Usually because the Instrumentation and Vocal slots are empty while Genre and Mood are overloaded with adjectives. Name two or three specific instruments and one concrete vocal direction, and add timestamps for structure.
What languages can Lyria 3 Pro vocals be in? Eight: English, German, Spanish, French, Hindi, Japanese, Korean, and Portuguese. Name the language explicitly in the vocal slot, and describe two singers with two languages for a bilingual duet.
Bottom Line
Better Lyria 3 prompts come from two habits, not from longer adjective lists: fill every slot of the official formula with one specific detail, and use timestamps to direct the structure section by section. The formula sets the sound; the timestamps set the shape. Do both and the model stops handing you three-minute loops and starts handing you songs.
The fastest way to internalize it takes about two minutes: pick one of the four examples above, change a couple of details to match a song you want, and run it through the Lyria 3 Pro generator. Test the plain formula at 30 seconds first, then add your timestamp skeleton and generate the full track.
Sources
Primary sources for the factual claims in this article. Note: Lyria 3 Pro is partly in public preview — capabilities may change after publication (July 2026).
- Google Cloud: Ultimate prompting guide for Lyria 3 Pro — prompt formula, flags, timestamp technique, languages
- Google: Lyria 3 Pro announcement — 3-minute songs, structure, vocals and lyrics
- Google AI for Developers: Generate music with Lyria 3 — prompting and generation via the API
- Google DeepMind: Lyria 3 model card — 48 kHz output, 8 vocal languages, SynthID
- Google Cloud: Lyria 3 and Lyria 3 Pro on Vertex AI — image and PDF inputs, C2PA credentials
- Google DeepMind: Lyria model page — model family and capabilities
