Google Lyria 3 Pro: cinematic instrumental music from text

Vocal AI music gets most of the attention, but the jobs that pay the most in video production are almost always instrumental: the tension cue under the trailer cut, the ambient bed behind the podcast, the four-second sting that opens a brand reel. Google Lyria 3 Pro was built for exactly this — coherent, polished instrumental compositions that behave like real scored music, not looped samples.

Lyria 3 Pro is part of Google DeepMind's music model family, refined after the February 2026 launch of Lyria 3 (30-second tracks in the Gemini app) to handle tracks up to roughly three minutes with richer creative control. On OmniArt it runs at 20 credits per track — the most affordable music model on the platform — and accepts prompts up to 5,000 characters, the longest music prompt budget available. That combination makes it the go-to choice when you need a polished instrumental score without spending your credit budget on a first draft.

What makes Lyria 3 Pro different from other music models

Lyria 3 Pro is an instrumental-only model. It does not accept lyrics and does not generate vocals. That constraint is also its strength: every parameter of the model is tuned toward compositional coherence rather than vocal performance. The result is music that holds together across its full length — dynamics that build and release, transitions between sections that feel intentional, and arrangements that stay in genre rather than drifting.

The model responds well to structural direction. You can describe an intro, a verse-level development, a climax, and a resolution in plain prose, and the model follows that arc in a way that vocal-first models — which are optimising for a lyric-delivery path — often miss. Lyria 3 Pro outputs are watermarked with SynthID in Google's own deployments; on OmniArt the output is delivered as a standard audio file.

Track length is decided automatically by the model based on the prompt content and implied structure. There is no duration control on OmniArt; describe the arc you want and the model will produce a length that fits it.

How to write a Lyria 3 Pro prompt

The 5,000-character prompt window is an invitation to be specific. Lyria 3 Pro rewards detailed, layered descriptions — a one-sentence prompt produces a usable draft, but a paragraph-level prompt that covers mood, instrumentation, dynamics, and structure produces something closer to a directed score.

Build your prompt from five dimensions:

Mood and atmosphere Name the emotional register and the setting. "Epic and triumphant" and "intimate and melancholic" produce very different outputs. Be specific: "quiet dread with a slow build to catharsis" gives the model a trajectory, not just a state.

Instrumentation List the sounds that anchor the piece. The model responds to concrete instrument names: solo cello, French horn section, plucked harp, low brass, ethereal choir pads, pizzicato strings, electric piano, prepared piano, synth pads, solo flute. Avoid vague terms like "orchestra" when you can name the instruments.

Dynamics and energy arc Describe the shape of the piece over time. "Starts sparse and grows to a full orchestral swell" or "stays minimal and tense throughout with a single release at the end" gives the model structural direction.

Section structure Lyria 3 Pro understands compositional language. You can describe sections in prose: "a quiet, single-instrument intro; a first theme that develops with layered strings; a mid-piece climax with full ensemble; a return to the solo theme for the outro." The model maps this to an actual arrangement.

Tempo and rhythm Describe pace in language rather than BPM. "Slow-burning, about 60 bpm feel," "driving, urgent triplet feel," "loose rubato, as if for a scene with no fixed tempo."

Three complete worked prompts

Example 1: cinematic trailer cue

This prompt targets the most common use case — a tension-building score for a video trailer, product launch, or short film.

Prompt:

Cinematic orchestral score, urgent and escalating, for a science fiction trailer. Opens with a solo cello playing a simple five-note motif against a low sustained drone. After eight bars, low brass and timpani enter underneath, the motif fragmenting into shorter phrases. Strings build through the middle section in staggered entries — first violins, then violas, then cellos — adding density each time. At the climax, full brass, massed strings, and a snare tattoo drive the peak. The final section strips back to the original cello motif, now harmonised by a single French horn, fading to silence. Throughout, a soft synth texture provides a futuristic underlay. Tempo: driving, around 120 bpm feel. No electronics other than the synth pad bed.

Why it works: the prompt names every instrument, maps the structural arc section by section, specifies the thematic development (the five-note motif recurring), and gives a clear dynamic shape from sparse to massive to sparse again. Lyria 3 Pro can follow this as a compositional brief.

Example 2: lo-fi ambient study bed

This prompt targets a different use case — a long-form ambient background, suitable for study, podcast underlay, or a creator's "always-on" background track.

Prompt:

Ambient lo-fi study music, calm and slightly nostalgic, for a long listening session. The piece should feel like a room on a rainy afternoon — warm, unhurried, and slightly worn. Instrumentation: close-mic electric piano as the main voice, with a steady soft bass note on each bar, brushed drums sitting very low in the texture, occasional low-volume guitar plucks, and a vinyl crackle texture underneath. No melodic development or dramatic arc — the goal is steady, meditative consistency. Tempo: loose and slow, around 70 bpm feel. Dynamics stay constant throughout with no builds or drops. Soft reverb on the piano. Occasional very brief pauses in the piano line, as if the player is thinking.

Why it works: this prompt explicitly instructs the model NOT to develop dynamically — a useful negative instruction for ambient work. It names the vinyl texture as a sonic element, specifies the bpm range, and describes the "feel" of the piece in scene-setting language (rainy afternoon room) alongside the technical instrumentation.

Example 3: corporate brand sting

This prompt targets a short, identity-level piece — a 10–20 second sting suitable for a brand intro, social media header, or product reveal transition.

Prompt:

Short brand identity sting, modern and polished, for a technology company in the productivity software space. Tone: confident, clean, forward-looking — not aggressive or triumphant, but assured. Instrumentation: a light synth pad chord, a single piano note on the downbeat, a short rising string figure, and a soft metallic chime that rings through the end. The total arc should be brief: opening chord — short melodic phrase — single resonant close. Minimal. Everything sits in the mid-to-high register; no heavy bass or low-end weight. The piece ends on a clear, sustained note rather than fading out. It should feel like a logo sound, not a full musical idea.

Why it works: the prompt defines the emotional role of the music (logo sound, not a full idea), explicitly restricts the length and arc, names the closing behaviour ("ends on a clear, sustained note"), and avoids genre labels in favour of character descriptions ("confident, clean, forward-looking"). This is the kind of brief a real composer would recognise.

Using Lyria 3 Pro for video scoring

The practical case for Lyria 3 Pro over stock libraries is that you control the brief instead of searching for an approximate match. Every prompt above can be adjusted to fit a specific scene length, brand character, or tonal requirement — regenerate until it's right.

On OmniArt, audio and video generation sit in the same workspace. A workflow that combines both modalities looks like this:

1. Generate or import your video clip using one of OmniArt's video models — Veo 3.1, Sora, Kling, or others.
2. Open the Audio workspace and select the Music tab.
3. Write a Lyria 3 Pro prompt describing the emotional content of the scene — not just the music in the abstract.
4. Generate two or three takes and audition them against picture. The model's output at this quality level usually lands within one or two iterations.
5. Pair with a sound effects pass from the same workspace for a complete soundbed. See the full audio model overview for SFX and ambience options.

Genres and use cases where Lyria 3 Pro particularly shines

• Cinematic and orchestral — full dynamic range, real instrument timbres, structural builds and releases
• Ambient and atmospheric — long-form consistency, textural sophistication, subtle variation
• Corporate and brand — clean, professional character without the "stock music" energy
• Documentary and narrative — emotional support without melodrama; useful for pacing interviews
• Game and interactive — loopable compositions and neutral-energy beds that don't fatigue after repeated listening

Lyria 3 Pro vs MiniMax Music 2.6 vs ElevenLabs Music

OmniArt's audio workspace includes three music models. The right choice depends on whether you need vocals and what the budget is.

Pick Lyria 3 Pro when: you need a high-quality instrumental score and want to keep costs down. The 20-credit price makes it practical for multiple-take iteration. The 5,000-character prompt window rewards detailed briefs.

Pick MiniMax Music 2.6 when: you need a vocal performance, want to provide your own lyrics, or are working to a tighter credit budget and want the default vocal-capable model.

Pick ElevenLabs Music when: section structure and arrangement fidelity are the priority, or when budget is less of a constraint than a specific sonic outcome.

For a side-by-side look at the vocal-first workflow with MiniMax Music 2.6, see MiniMax Music 2.6: generate full songs with lyrics.

Getting started on OmniArt

Open the audio workspace and select the Music tab. Choose Google Lyria 3 Pro, write a prompt that covers mood, instrumentation, dynamics, and section structure — using the examples above as a starting point — and generate your first track.

At 20 credits a take, Lyria 3 Pro makes it practical to generate five or six variations of a brief and pick the one that fits the picture. Start with the trailer cue prompt above, adjust the instrumentation and scene description to match your project, and generate. The gap between brief and usable result is usually one or two passes.