This website is not affiliated with Suno. For official Suno services, please visit suno.com

Google Lyria 3 Pro: The Complete Developer Guide to Google's Best AI Music Model

Published 2026-05-20 · ~10 min read

Google Lyria 3 Pro is the current flagship music-generation model from Google DeepMind, capable of producing up to 3 minutes (180 seconds) of studio-quality audio per call with multilingual vocals, custom lyrics, and image-conditioned composition. It is the model that powers Google's Flow Music, the YouTube Shorts Dream Track feature, and the music-generation surfaces inside Gemini. Every output is invisibly watermarked with Google's SynthID for provenance attestation. It is exposed to developers today through the AI Music API /v1/producer/* endpoints (formerly branded as Producer AI).

"Lyria 3 Pro is our most advanced music generation model yet."

— Google DeepMind, Lyria technologies page

This guide walks through everything a developer needs to know to ship with Lyria 3 Pro today: what it can do, how it differs from Lyria 2 and from Suno V5, how to call it via API, real code, real pricing, and where to draw the boundary between Lyria and other models for different jobs.

What is Google Lyria 3 Pro?

Lyria is Google DeepMind's text-and-image-to-music model family. The "3 Pro" suffix marks the current flagship; Google explicitly describes it as their "most advanced music generation model yet" on the DeepMind technologies page. It's distinct from:

  • Lyria 3 RealTime, a low-latency variant tuned for live performance and real-time DJ-style workflows
  • Magenta RealTime, DeepMind's open-source companion model for research and on-device use
  • Lyria 2, the previous flagship; still good, but Lyria 3 Pro generates longer tracks at higher fidelity and adds the image-to-music modality

The model is what powers Google Flow Music, the music generation feature inside Google Vids, and the YouTube Shorts Dream Track integration. Until recently it was only available to Google products. The "production-ready API" surfacing it to third-party developers is new.

What's new in Lyria 3 Pro

The headline upgrades over Lyria 2:

  1. Longer tracks: up to 3 minutes per call. Lyria 2 capped at roughly 90 seconds. Lyria 3 Pro doubles that, which is what unlocks full-song generation (verse / chorus / bridge / outro) without stitching multiple jobs together.

  2. Multilingual vocals. Lyria 3 Pro generates singing voices across multiple languages with natural pronunciation, not just transliterated lyrics with an English accent. This was the biggest practical limitation of every previous version.

  3. Image-to-music composition. Upload a reference image; the model reads the visual mood (color palette, subject, energy) and composes a track that fits. Useful for app builders generating soundtracks from user-uploaded photos or video keyframes.

  4. Fine-grained acoustic control. Tempo, dynamics, instrumentation, key, and vocal style are now structured parameters rather than freeform prompt text. The model reliably follows numeric tempo (e.g. 90 BPM) and discrete style tokens.

  5. SynthID watermarking by default. Every output ships with Google's invisible audio watermark. Useful for provenance attestation in regulated industries (advertising, news, social platforms).

  6. Genre breadth. The DeepMind page calls out "pop to funk to Motown", but in practice, the model handles cinematic, lo-fi, classical, hip-hop, EDM, and folk equally well. Per-genre prompt engineering is unnecessary.

Lyria 3 Pro capabilities at a glance

| Dimension | Lyria 3 Pro | | --- | --- | | Max duration | 3 minutes (180 seconds) — 2× Lyria 2's ~90-second ceiling | | Audio quality | Studio-grade, 44.1 kHz / 16-bit, suitable for production master | | Languages (vocals) | Multiple (multilingual) — English, Spanish, French, German, Japanese, Korean, Chinese verified | | Input modalities | Text, image, lyrics | | Output | High-fidelity audio (MP3 + optional WAV) + SynthID watermark | | Control parameters | Tempo (numeric BPM), dynamics, key, instrumentation, vocal style, theme | | Editing endpoints | Cover (instruction-based), vocal swap, sound swap, section replace | | Latency | p50 ≈ 45 seconds, p95 ≈ 90 seconds end-to-end via async API | | Commercial use | Yes, royalty-free license shipped with every track via AI Music API |

How to access Google Lyria 3 Pro via API

There's no direct public API from Google DeepMind itself. Lyria is exposed inside Google's own products (Gemini, YouTube, Flow). To call it from your own app, you need an API gateway that bridges the gap.

That's what AI Music API does. We're a developer-facing API that routes Lyria 3 Pro requests, hosts the resulting audio on our own CDN, handles async job callbacks, and bills per generation. Sign up at aimusicapi.ai/en/signin, generate a bearer token from the dashboard, and you're done with setup.

The minimal request

A single POST creates a Lyria 3 Pro track:

curl -X POST https://api.aimusicapi.ai/v1/producer/create \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "lyria-3-pro",
    "lyrics": "[Verse]\\nWalking through the city lights tonight\\n[Chorus]\\nFeel the rhythm of the streets",
    "style": "cinematic pop, female vocal, 90 BPM",
    "title": "Night Drive",
    "duration": 180,
    "instrumental": false,
    "hook_url": "https://your-app.example/webhooks/music"
  }'

You get back a task_id immediately. Your webhook receives a callback with the audio URL when generation completes, typically under a minute.

What each field does

  • model: set to lyria-3-pro to use the current flagship. You can swap to lyria-3-realtime for low-latency, or to any of the Suno models for a different sound.
  • lyrics: your lyrics, formatted with [Verse], [Chorus], [Bridge] tags. Skip this and pass instrumental: true to generate purely instrumental tracks.
  • style: natural-language style prompt. Genre + instrumentation + vocal hints + numeric tempo all work.
  • title: used as the track filename and inside metadata.
  • duration: seconds; max 180 for Lyria 3 Pro.
  • instrumental: true to skip vocals entirely.
  • hook_url: your webhook endpoint. Receives a JSON POST with the final audio URL when the job completes.

Polling alternative

If you can't host a webhook, poll the task endpoint:

curl https://api.aimusicapi.ai/v1/producer/task/$TASK_ID \
  -H "Authorization: Bearer YOUR_API_KEY"

Returns status: "PENDING" | "PROCESSING" | "SUCCESS" | "FAILED" plus the audio URL on success.

Cover, remix, and section editing

Lyria 3 Pro isn't just for text-to-music. The same backend exposes editing endpoints driven by natural-language instructions:

  • Cover: Provide a source track + an instruction ("convert to jazz piano") + a strength parameter (0–1). The model re-renders the song in the target style while preserving lyrics and structure.
  • Vocal swap: Replace the vocal performance while keeping the original backing track. Useful for trying different singers on the same song.
  • Sound swap: Inverse of vocal swap: keep the vocals, re-render the instrumental.
  • Section replace: Regenerate a specific time window (e.g. 20s–120s) while keeping the rest of the track intact.

All four use the same model: "lyria-3-pro" parameter; the endpoint determines the operation. See the Producer API docs for the full request shapes.

Google Lyria 3 Pro vs Suno V5

The most common question developers ask: which model should I choose?

The honest answer is both, depending on the job. Here's how they split:

| | Google Lyria 3 Pro | Suno V5 | | --- | --- | --- | | Max duration | 3 minutes | ~4 minutes (with extend) | | Vocal style | Multilingual, fine-grained vocal control | Multilingual, persona-based vocals | | Image input | Yes (image-to-music) | No | | Audio input | Cover via instruction | Cover, extend, MIDI alignment | | Watermark | SynthID (Google) | Suno fingerprint | | Stems | Not exposed | Yes (basic + full separation) | | Best for | Long-form, controlled, brand-safe music | Fast turn-around, broad genre coverage, stem-aware workflows | | Pricing (via AI Music API) | $0.08 / generation | $0.08 / generation |

Use Lyria 3 Pro when: you need long-form tracks, controlled vocal style, image-driven composition, or SynthID provenance attestation for brand-safe distribution.

Use Suno V5 when: you need stem separation (vocals / drums / bass / other), broader Western pop genre coverage, or the fastest possible turn-around.

The good news: through AI Music API, both models live behind the same account, the same billing, and the same REST surface. Switch the model field per request.

Production use cases

Real shipping integrations using Lyria 3 Pro via our API:

  1. Background music inside video editing tools, Generate a soundtrack matching a video's mood (image-to-music from a keyframe) and time-align it to the cut.
  2. Indie game soundtracks, Lyria 3 Pro's 3-minute output is enough for full level music; loop a track via in-engine logic and ship without licensing fees.
  3. Personalized music inside creator apps, "Birthday song for [name]", "wedding playlist matching this photo album", user-prompted, brand-safe (SynthID), production-licensed.
  4. Marketing audio at scale, Programmatic ad jingles, sonic logos, podcast intros. Each Lyria call is $0.08, under-the-noise of any reasonable creative budget.
  5. Music-as-a-feature inside SaaS, Generate-and-export functionality inside your own UI, white-labeled.

Pricing

Lyria 3 Pro generations through AI Music API are billed at $0.08 per generation (10 credits). New accounts get 30 free credits on signup, enough to test the full flow before committing a payment method.

For higher-volume teams, monthly scale plans bundle credits with raised concurrency limits, priority support, and a single B2B invoice instead of usage-based top-ups. See the full pricing matrix at aimusicapi.ai/en/pricing.

There are no per-project licensing fees, no royalty deductions, and no gating of commercial use. Every generated track ships with a commercial license, use it in apps, games, ads, and shipped products without further negotiation.

Getting started

The full onboarding flow is three steps:

  1. Sign up at aimusicapi.ai/en/signin. Email + password or Google OAuth.
  2. Grab your API key from the dashboard. Keep it server-side; never expose it in browser-shipped code.
  3. POST your first request using the curl example above. Set the webhook URL or poll the task endpoint.

Total time to first generated track: typically under 5 minutes.

FAQ

Is Google Lyria 3 Pro free to use?

The model itself is exposed through paid surfaces (Gemini, Google Vids, AI Music API). There's no free direct-API access from Google. AI Music API gives every new account 30 free credits (~3 generations) to evaluate Lyria 3 Pro before paying.

What languages do Lyria 3 Pro vocals support?

Multiple, Google describes it as "multilingual" without publishing a fixed list. In practice, English, Spanish, French, German, Japanese, Korean, and Chinese all work reliably. Specify the language in the lyrics prompt or via style hints.

Can I commercially use Lyria 3 Pro output?

Yes, through AI Music API, every track ships with a commercial license. The SynthID watermark is invisible to listeners and serves as provenance attestation rather than restriction.

How long does generation take?

Typically under a minute end-to-end via the async REST API. Long-form 3-minute tracks may take slightly longer than shorter ones, but generation is parallelized across our infrastructure to keep latency consistent.

Can I edit a Lyria 3 Pro track after generation?

Yes, the cover, vocal swap, sound swap, and section replace endpoints accept a previously-generated task_id and modify the track based on natural-language instructions plus a strength parameter.

How does Lyria 3 Pro compare to ElevenLabs Music?

Both target professional-grade music generation. Lyria 3 Pro emphasizes long-form tracks and image-to-music composition; ElevenLabs Music emphasizes vocal-cloning-style sonic control. Through AI Music API we currently ship Lyria 3 Pro, Suno V5, and Riffusion variants; ElevenLabs Music is available through our Eleven Music API page.

Build with Google Lyria 3 Pro today

You've read this far. The next step is the curl request. Get 30 free credits and ship your first Lyria-powered feature this afternoon.

ShareXLinkedInReddit