Modern pop
Full-length vocal song
Reserved for a complete arrangement with intro, verses, choruses and a bridge inside the official length ceiling.
Vocals · ~3:00Audio coming soon
LYRIA 3 PRO · INDEPENDENT MULTI-MODEL STUDIO
Build structure-aware full songs with Lyria 3 Pro
Describe the genre, mood, instrumentation, tempo, vocals and lyrics you want, lay out the sections with timestamps, and render a complete track of up to 184 seconds — with copy-ready prompts, a walkthrough and a live point estimate.
- Up to 184 seconds
- Duration, BPM and intensity
- Text, image or PDF input
- 8 vocal languages
- Live point estimate
Nano Banana Pro is an independent multi-model AI creation platform. It is not the official website of Google, Google DeepMind, Gemini, Flow Music or Vertex AI, and has no affiliation with or endorsement from them.
LIVE WORKFLOWGoogle FlowMusic Lyria 3 Pro
Up to 184 secondsDuration, BPM and intensityText, image or PDF input
01
Lyria 3 Pro online music workspace
Available fields, model status and the estimated point cost come from the live catalogue — please confirm them before you submit.
Google FlowMusic Lyria 3 Pro
AI Music Studio
Estimated cost — points
Model
Loading live model settings…
Generated audio
♫Your generated audio will appear here.
02 · MUSIC DEMOS
Real audio coming soonLyria 3 Pro demo slots
These slots are reserved for finished tracks once real audio is published. Until then every card is explicitly labelled as a placeholder, and no fake player is shown.
Cinematic
Scored film sequence
Reserved for a cue whose dynamics track scene changes, written with timed segments.
Instrumental · ~1:30Audio coming soon
Commercial
Brand audio tag
Reserved for a compact hook and a memorable brand landing point.
Vocals · ~0:30Audio coming soon
Lo-fi
Lo-fi focus loop
Reserved for a steady groove with soft texture that loops cleanly.
Instrumental · ~2:00Audio coming soon
Multimodal
Storyboard-to-score
Reserved for a track generated from reference images so the mood follows the visuals.
Instrumental · ~0:45Audio coming soon
Timestamped
Genre-shift collage
Reserved for a track that changes genre on a schedule using time-aligned segments.
Vocals · ~2:15Audio coming soon
03 · MODEL OVERVIEW
Lyria 3 Pro at a glance
Lyria 3 Pro is Google’s full-song variant in the Lyria 3 music generation family, released in public preview on 25 March 2026 under the API model ID lyria-3-pro-preview. Google’s launch post says it "unlocks longer tracks with structural awareness", and the developer announcement describes it as the premier model for full-length song generation with professional-grade structural awareness. The official blog puts the ceiling at roughly three minutes; the published model card gives the precise figure of 184 seconds.
Google’s own prompting guide lists three strengths: structural control, so you can ask for intros, verses, choruses and bridges; high-quality audio, described as high-fidelity stereo; and precision control through timed lyrics, descriptive tempo conditioning and multimodal inputs. The same guide gives the official prompt framework, the ability to lift your own lyrics with a Lyrics: prefix, vocal generation in eight languages, and the ability to run the model from text, a PDF or up to 10 reference images.
This page is an independent multi-model workspace rather than a Google product. Everything above describes the model as Google documents it. Which fields this site exposes, what each run costs and how long it actually takes are read from the live catalogue at submit time, so treat the workspace panel as the source of truth for what you can run right now.
- Model ID
- lyria-3-pro-preview (public preview)
- Released
- 25 March 2026
- Length ceiling
- 184 seconds; Google blogs describe it as up to about 3 minutes
- Audio output
- 44.1 kHz, 192 kbps, audio/mp3
- Structure controls
- Duration, BPM and intensity
- Vocal languages
- 8: English, German, Spanish, French, Hindi, Japanese, Korean, Portuguese
04 · CAPABILITIES
What Lyria 3 Pro can generate
Capability notes below are summarised from Google’s own model card, launch posts and prompting guide. Which fields this site exposes, and what a run costs, is shown live in the workspace panel.
Full-length songs
The model is positioned for complete compositions rather than clips, with an official ceiling of 184 seconds.
Structural awareness
Prompt for intros, verses, choruses and bridges instead of hoping the arrangement arrives on its own.
Timed lyrics and segments
Lay out [mm:ss] markers so each section starts when you want it to, and let the track shift genre on a schedule.
Expressive vocals
Steer singer type, range, timbre and delivery, with multi-vocal generation across eight languages.
Your own lyrics
Write Lyrics: before the lines you want performed, and mark backing vocals inline where you need them.
Instrumental mode
Add the word instrumental to strip vocals out entirely and leave a clean soundbed.
Multimodal input
Beyond text, Google’s workflows accept a PDF or up to 10 reference images to set the emotional baseline.
Provenance built in
Google states all Lyria 3 and Lyria 3 Pro output carries a SynthID watermark and supports C2PA content credentials.
05 · USE CASES
Where Lyria 3 Pro earns its keep
Video and short-form scoring
Build a bed that follows the pacing of a clip, then use timed segments to hit the cuts.
Brand, ad and product audio
Sketch a jingle, campaign theme or sound direction for a launch before you commit production budget.
Songwriting drafts
Turn a lyric idea, a genre brief and a structure note into a full-length demo you can play for other people.
Podcast and show packaging
Produce intro, outro and transition beds that match the tone of a series.
Film and game concepts
Test the musical mood of a character, location, trailer or level at a length that survives a full scene.
Visual-to-music work
Hand over a storyboard, a mood board or a PDF brief and let the imagery set the direction of the score.
06 · PROMPT GUIDE
Copy-ready Lyria 3 Pro prompts
Google’s prompting guide recommends a structured prompt over a pile of adjectives. Each card below follows the official six-part framework and states its own length target, so you can copy one, change a single variable and compare results.
Official framework: [Genre and style] + [Mood] + [Instrumentation] + [Tempo and rhythm] + [Vocal style & language] + [Lyrics]
Bossa nova and R&B duet
A romantic fusion of classic Bossa Nova and modern R&B. The mood is intimate, warm and deeply affectionate. Features a gentle acoustic nylon-string guitar, warm electric piano chords and a crisp, laid-back modern hip-hop drum beat. A slow, swaying tempo. A vocal duet: a smooth male vocalist in English and a soft, breathy female vocalist in French. The lyrics are a love song about an undeniable cross-cultural connection.
- Tags
- Bossa nova, R&B, duet, slow tempo, nylon-string guitar
- Mode
- Vocals · 165 sec
- Exclude
- harsh synth, shouted vocals, drum solo
Gospel to hip-hop genre shift
[00:00] Begin with a large gospel choir singing an uplifting harmony about being kind to yourself. [00:15] A heavy modern hip-hop beat and a deep bassline drop in at the choir’s energy. [00:35] A male lead vocalist starts a confident rap verse while the choir punctuates his lines. [01:10] Move into a triumphant chorus at full volume. [01:50] Strip back to a warm organ for a quiet, emotional bridge. [02:30] Return at maximum energy for a final chorus and end on a sustained choir chord.
- Tags
- Gospel choir, hip-hop, timestamped, genre shift
- Mode
- Vocals · 180 sec
- Exclude
- lo-fi percussion, whispered delivery, abrupt cutoff
Dawn cinematic cue
A cinematic orchestral cue about the last hour before sunrise. It opens on a lone piano and low strings, then adds horns, tuned percussion and a wide string section. The mood moves from restraint to expectation and finally release, with a natural dynamic arc and a tail that leaves room for a picture cut. No vocals anywhere in the track.
- Tags
- Cinematic, orchestral, gradual build, hopeful
- Mode
- Instrumental · 90 sec
- Exclude
- vocals, EDM drop, distorted guitar
Rainy night study loop
A warm lo-fi hip-hop instrumental for late-night reading, around 72 BPM. Soft jazz piano chords, rounded bass, light vinyl texture and a restrained drum loop. Keep the arrangement sparse and even, avoid any dramatic climax, and make the end loop back into the beginning cleanly. Instrumental.
- Tags
- Lo-fi hip-hop, jazz piano, 72 BPM, instrumental
- Mode
- Instrumental · 120 sec
- Exclude
- vocals, bright lead, heavy drums
Bright J-pop with male tenor
An upbeat, high-energy J-pop track with bright sparkling synths, electric guitar and a driving bassline. A clear, expressive male tenor sings in Japanese, with a fast-paced melodic delivery and a polished, sweet texture. The chorus should open up wide and repeat a hook that is easy to sing back.
- Tags
- J-pop, male tenor, Japanese, bright synths
- Mode
- Vocals · 60 sec
- Exclude
- slow ballad tempo, raspy growl, orchestral only
Thirty-second brand tag
A polished commercial pop tag, around 118 BPM, built on bright synths, a tight drum kit and a short English vocal line. Establish the hook inside the first five seconds, keep the middle sparse so a voiceover can sit on top, and land the final brand line on a clean two-beat stop.
- Tags
- Commercial pop, 118 BPM, hook, voiceover space
- Mode
- Vocals · 30 sec
- Exclude
- long intro, dark mood, complex solo
Storyboard to score
A modern documentary score whose lyrics and mood match the story told by the attached images. Intimate and observant in the opening, opening up to something more confident as the images move outdoors, and closing on a quiet sustained pad. Instrumental, with no lead vocal and no percussion hit in the last ten seconds.
- Tags
- Documentary, image-to-music, intimate, sustained pad
- Mode
- Instrumental · 45 sec
- Exclude
- vocals, aggressive percussion, comedic tone
07 · TUTORIAL
A Lyria 3 Pro walkthrough
Six steps that take you from a rough idea to a full-length draft you can actually evaluate.
- 01
Decide the deliverable and the length
Write down whether you need a full song, a scored scene or a short tag, and the target length. Anything up to 184 seconds is inside the official ceiling; longer ideas need to be cut down before you prompt.
- 02
Fill in the six-part framework
Work through genre and style, mood, instrumentation, tempo and rhythm, vocal style and language, then lyrics. Outlining the prompt in that order stops you from burying the important detail at the end.
- 03
Choose instrumental or vocal
Add the word instrumental if you want a soundbed. If you want singing, name the singer type, range, timbre, delivery and language rather than leaving the performance to chance.
- 04
Lay out the structure with timestamps
For anything with a story arc, write [mm:ss] markers and describe what happens in each segment. This is the fastest way to control where a chorus lands or where a genre shift happens.
- 05
Add reference material if the mood is hard to describe
Google’s workflows accept a PDF or up to 10 reference images. Attach a storyboard or a mood board when words are not getting you close enough to the feeling you want.
- 06
Generate, listen, change one thing
Confirm the live point estimate before you submit. When you listen back, change a single variable per run — tempo, instrumentation or one section — so you can tell which edit actually improved the track.
08 · COMPARE
Lyria 3 Pro vs Lyria 3 Clip
Lyria 3 Pro and Lyria 3 Clip are the two variants Google publishes side by side on the same model card, released on the same day. Choosing between them is mostly a question of how long the finished audio has to be, not which one sounds better.
| Item | Lyria 3 Pro | Lyria 3 Clip |
|---|---|---|
| Song generation | Full song | 30-second clips |
| Maximum length | 184 seconds, described in Google blogs as about 3 minutes | 30 seconds |
| Structure controls | Duration, BPM and intensity | BPM and intensity only |
| Strongest at | Complete arrangements with intros, verses, choruses and bridges | Fast iteration, background loops and social clips |
| Google’s positioning | Premier model for full-length song generation | Optimised for speed and high-volume requests |
| API model ID | lyria-3-pro-preview | lyria-3-clip-preview |
| Choose it when | The track has to stand on its own as a finished piece | You are still exploring and want to test volume before committing |
Watermarks, rights and platform boundaries
Google states that every track produced by Lyria 3 and Lyria 3 Pro carries a SynthID watermark and supports C2PA content credentials, which keeps a verifiable record of how the audio was made even after editing. Provenance is not the same thing as permission: a watermark tells people the audio came from an AI model, it does not grant you a licence to use it.
Two official limits are worth knowing before you plan a project. The model card lists negative prompting as not supported, so treat any exclude field you see in a workspace as that platform’s own convenience rather than a Lyria 3 Pro capability. The supported vocal languages are English, German, Spanish, French, Hindi, Japanese, Korean and Portuguese, so Mandarin singing is outside what Google currently documents.
Google also states that Lyria 3 and Gemini do not mimic artists, and that if a prompt names a creator the model takes it only as broad inspiration. That protection sits with the model owner, not with you: if you supply lyrics, images, trademarks or a likeness, you are still responsible for having the right to use them.
10 · FAQ
Lyria 3 Pro questions, answered
What is Lyria 3 Pro?
Lyria 3 Pro is Google’s full-song variant in the Lyria 3 music generation family, published in public preview on 25 March 2026. Google describes it as the premier model for full-length song generation with professional-grade structural awareness, so it follows prompts for intros, verses, choruses and bridges.
How long can Lyria 3 Pro tracks be?
The published model card gives a maximum audio clip length of 184 seconds. Google’s launch blogs round this to about three minutes. Shorter targets are fully supported, so most prompts should simply state the length the project needs.
Is Lyria 3 Pro the same as Lyria 3 Clip?
No. They are two variants on the same model card. Lyria 3 Pro generates full songs up to 184 seconds and exposes duration, BPM and intensity controls. Lyria 3 Clip produces 30-second clips and is positioned for speed, rapid prototyping and high-volume requests.
How is Lyria 3 Pro different from Lyria 3.5?
Lyria 3 Pro arrived first, in March 2026, and is positioned around structural awareness for full-length songs. Google describes Lyria 3.5 as a later model with advances in musicality, lyrics and vocal quality. Google does not publish a field-by-field comparison of the two, so this page does not invent one.
Does Lyria 3 Pro support negative prompts?
Google’s model card lists negative prompting as not supported for this model. An exclude or negative field you see inside a third-party workspace is that platform’s own input handling, not a documented Lyria 3 Pro control, so do not rely on it to remove an element you dislike.
Can Lyria 3 Pro sing in Mandarin Chinese?
Not according to the current official documentation. Google lists eight supported vocal languages — English, German, Spanish, French, Hindi, Japanese, Korean and Portuguese — and Mandarin is not among them. You can still write Chinese-language prompts and describe a Chinese vocal direction, but Chinese singing is not a documented capability.
How do I write a Lyria 3 Pro prompt?
Use Google’s six-part framework in order: genre and style, mood, instrumentation, tempo and rhythm, vocal style and language, then lyrics. Add the word instrumental if you want no vocals. The more specific each part is, the more of the result you actually control.
Can I use my own lyrics?
Yes. Google’s guide says to write Lyrics: before the lines you want performed. If you want backing singers to echo the main vocal, note that in the prompt as well. Leaving the Lyrics: block out and describing a theme instead lets the model write the words for you.
Can I give Lyria 3 Pro an image or a document?
Yes. The prompting guide describes generating music from text, from a PDF, or from up to 10 reference images. Attaching visuals or a brief PDF is useful when the mood you want is easier to show than to describe in words.
Do Lyria 3 Pro tracks carry an AI watermark?
Google states that all Lyria 3 and Lyria 3 Pro output is embedded with a SynthID watermark and supports C2PA content credentials. The watermark is designed to survive editing so the audio can still be identified as AI-generated afterwards.
Can I use Lyria 3 Pro music commercially?
You cannot answer that from the model name alone. Check this site’s terms, the rules of whichever model supply you actually use, where your source material came from, and the law that applies to you. Being allowed to use a track commercially does not automatically give you exclusive rights in it.
Is this the official Google website for Lyria 3 Pro?
No. Nano Banana Pro is an independent multi-model creation platform. It is not affiliated with, operated by or endorsed by Google, Google DeepMind, Gemini, Flow Music or Vertex AI, and official Google terms and pricing do not automatically apply here.
Keep the workflow moving
CREATE NEXT
Back to the workspaceTurn a structure you can describe into a song you can hear
Start from a use case, a genre and a length, lay out the sections you want, then refine one variable at a time inside the live workspace.