Google Lyria 3 Pro: A Developer's Guide

What is Google Lyria 3 Pro?

Google Lyria 3 Pro is the production release of Google DeepMind's text-to-music model family. It accepts a natural-language prompt and optional lyrics, and returns a studio-quality audio clip with synthesized vocals, instrumentation, or both. Lyria 3 Pro is the successor to the earlier Lyria 3 preview tier and is positioned alongside Suno v5 and ElevenLabs Music as one of the top three production text-to-music systems available in 2026.

Three properties define Lyria 3 Pro from a developer's point of view:

1. Speed. A typical generation finishes in around 30 seconds. That is roughly 2x to 4x faster than Suno v5 and makes interactive product flows (preview-and-iterate, in-app generation) viable.
2. Quality consistency. Output stays in a tight band across genre, tempo, and arrangement. There's less "coin-flip" variance than earlier text-to-music models, which matters when you ship results directly into user-facing products.
3. Operation breadth. Lyria 3 Pro supports create, extend, replace, and cover through a single endpoint. You can produce a track, lengthen it, swap out a chorus, or remix it in a new style without switching APIs.

How to call Lyria 3 Pro in 5 minutes

Lyria 3 Pro is exposed through MusicAPI's Producer API at /api/v1/producer/*. The integration path is the same regardless of language : issue an HTTPS POST with an API key, receive a task_id, then poll or use a webhook for the finished clip.

Step 1 · Get a key

Sign up at musicapi.ai. New accounts receive 75 free credits, enough for six Lyria 3 Pro generations (12 credits each). No credit card required.

Step 2 · Submit a generation request

# Bash / curl
curl -X POST https://api.musicapi.ai/api/v1/producer/create \
  -H "Authorization: Bearer $MUSICAPI_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "task_type": "create_music",
    "sound": "cinematic orchestral trailer with powerful choirs",
    "title": "Trailer Cue",
    "length": 60
  }'

# Response (immediate, ~50ms):
# { "message": "success", "task_id": "29906f02-2366-4218-..." }

The length field bounds the clip to 60 seconds. Lyria 3 Pro accepts 1 to 240 seconds per call. Provide lyrics alongside sound if you want sung vocals: the model handles both literal and free-form lyric structure.

Step 3 · Wait for the clip

# Poll once the task has had ~30 seconds to render
curl -H "Authorization: Bearer $MUSICAPI_KEY" \
  https://api.musicapi.ai/api/v1/producer/task/29906f02-2366-4218-...

# Successful response:
# {
#   "code": 200,
#   "data": [{
#     "clip_id": "abc123",
#     "title": "Trailer Cue",
#     "audio_url": "https://...m4a",
#     "wav_url":   "https://...wav",
#     "image_url": "https://....jpg",
#     "state": "succeeded"
#   }],
#   "message": "success"
# }

For production traffic, prefer webhooks over polling. Pass a webhook_url at submit time and MusicAPI will POST the same payload to your endpoint when the task completes. HMAC-SHA256 signing is supported through webhook_secret.

Step 4 · Build a flow

Use the returned clip_id with any of the other operations:

• extend_music: continue from a timestamp. Pass clip_id, starts_at, and a fresh instruction.
• replace_music: swap a window. Pass clip_id, starts_at, ends_at, and your replacement instruction.
• cover_music: reinterpret in a new style. Pass clip_id, instruction, and strength (0 to 1). Re-enabled in May 2026 after a brief gap.

What Lyria 3 Pro can actually do

Three capability axes, with concrete prompt examples that work:

1. Instrumental composition

Set lyrics empty and include explicit "instrumental, no vocals" cues in your sound prompt. Lyria responds to these cues better than the older make_instrumental flag (which is silently ignored: it's a legacy parameter from the previous Producer model platform).

Prompts that work:

• "instrumental, no vocals, peaceful ambient meditation with soft piano and gentle nature sounds"
• "instrumental, cinematic orchestral score with sweeping strings, powerful brass, and dramatic crescendos"
• "instrumental, downtempo electronic chillout with warm analog synths and deep sub bass"

2. Vocal composition with lyrics

Provide sound for style and lyrics for the actual words. Use [Verse], [Chorus], and [Bridge] markers: Lyria respects song structure when you mark it explicitly.

{
  "task_type": "create_music",
  "sound": "indie pop ballad with acoustic guitar and warm vocals",
  "lyrics": "[Verse 1]\nWalking down the empty street\nFeeling rhythm in my feet\n\n[Chorus]\nThis is the night we'll remember\nUnder the stars in November",
  "title": "November Night",
  "length": 90
}

3. Iterative editing

Once you have a clip, you don't need to start over to change one part. Use replace_music with precise starts_at and ends_at to swap a chorus or a vocal section. Use cover_music to reinterpret the whole song in a different genre with adjustable strength:

# Same source clip, three different covers at different intensities
{
  "task_type": "cover_music",
  "clip_id": "abc123",
  "instruction": "acoustic guitar version, intimate coffeehouse arrangement",
  "strength": 0.3   // stays close to source
}

{
  "task_type": "cover_music",
  "clip_id": "abc123",
  "instruction": "synthwave 80s with gated reverb drums",
  "strength": 0.85  // dramatic reinterpretation
}

Lyria 3 Pro vs Suno v5 vs ElevenLabs Music

All three are studio-grade text-to-music models with distinct strengths. The right choice depends on your use case.

• Lyria 3 Pro: fastest turnaround. Strongest for interactive product flows where users iterate. Cleanest dynamics across orchestral and electronic music. Use when speed and predictability matter.
• Suno v5: broadest stylistic range and most distinctive vocal styles. Stronger MIDI/stems ecosystem. Longer per-task ceiling (up to 8 minutes via extend). Use when you need genre-specific voice work or longer tracks in one shot.
• ElevenLabs Music: purpose-built for multi-lingual vocal control and structured arrangement. Smaller ecosystem of editing operations. Use when natural-language vocal control or non-English vocals are central.

All three are available through MusicAPI under one billing surface. The recommended pattern is to A/B route based on use case rather than pick one model and commit.

Pricing and economics

Lyria 3 Pro costs 12 credits per task on MusicAPI. Pricing does not vary by clip length, by instrumental vs vocal, or by prompt complexity. Upload tasks (uploading your own audio to use as a source for extend/replace/cover) and download tasks (exporting WAV) cost 2 credits each. Polling task status is free.

Failed upstream tasks are auto-refunded: if the model returns FAILED or the job times out, MusicAPI credits the cost back without a support ticket. This matters in production because text-to-music models occasionally fail on edge prompts and you don't want sporadic upstream issues to silently burn customer credits.

Best practices we've learned

1. Write prompts like a music director, not a search query. "upbeat pop" gives generic results. "mid-tempo synth-pop with a warm analog bass, gated reverb on the snare, and a soaring vocal hook" gives a finished song. The 10 to 30 word range hits the sweet spot.
2. Use explicit structure markers in lyrics. Lyria respects [Verse], [Chorus], [Bridge], [Drop], and [Outro] markers. Without them, song structure becomes a coin flip.
3. Set length explicitly for product flows. The default length is upstream-determined and varies. If your UI expects a 60-second clip, set length: 60.
4. Use instruction, not sound, for edits. For extend_music / replace_music / cover_music, editorial guidance is much clearer through an explicit instruction field than by hoping the model infers your intent from a sound prompt.
5. Use a seed when you A/B test. Same prompt + same seed = reproducible result. Useful for tuning prompt wording without confounding from generation randomness.
6. Always wire a webhook in production. Polling works for prototypes; webhooks scale better and free you from request-budget math.

Common questions

Next steps

• Google Lyria 3 Pro API landing page: full spec sheet, comparison table, and use cases.
• Producer API docs: complete endpoint reference, request/response schemas, and code samples.
• Playground: try Lyria 3 Pro in the browser, no signup required.
• Pricing: plans, credit ladder, and the 75-credit free tier.

Last updated 2026-05-20. The MusicAPI Producer API was migrated to Google Lyria 3 Pro in April 2026, with cover_music re-enabled in May 2026.