Lyria

Lyria Music Generation Skill

Generate high-fidelity 44.1 kHz stereo audio, full songs with structured lyrics, instrumental soundtracks, and thematic compositions from text prompts or visual reference images using Google's Lyria foundation models (lyria-3.5 and lyria-3-clip-preview).

Available scripts

• scripts/lyria.py: CLI tool for text-to-music and image-to-music generation, lyric export, and audio format conversion.
• scripts/test_lyria.py: Test suite verifying Lyria argument parsing and multimodal input limit enforcement.

Trigger Conditions

Activate this skill whenever the user asks to:

• Generate music, songs, soundtracks, game audio beds, or background music.
• Compose musical themes inspired by artwork, concept photos, or screenshots.
• Create 30-second audio loops or multi-minute structured songs with verses, choruses, and custom lyrics.
• Synthesize instrumental audio tracks ("Instrumental only, no vocals").

Model Selection & Comparison Matrix

Model Selection Guide

• Full Song Production: Use Lyria 3.5 (lyria-3.5, default) for complete tracks with timestamped transitions ([0:00 - 0:15] Intro...), multi-verse vocal delivery, or cinematic thematic development.
• Rapid Prototyping & Loops: Start with Lyria 3 Clip (clip) to iterate quickly on genre combinations, tempos, and 30-second seamless loops.

Deep Technical References

Consult dedicated reference cards in references/ for detailed audio parameters, timing controls, and API schemas:

• references/lyria-3-5.md: Flagship specifications for full-length songs, rich vocal synthesis, and arrangements.
• references/lyria-3-clip.md: Specifications for 30-second clips, loops, and rapid prototyping.
• references/lyria-3-pro.md: Specifications for legacy full-length preview model, timestamp controls, and scoring.
• references/README.md: Musical composition prompt reference and index.

Core Execution Workflows

1. CLI Execution via scripts/lyria.py

Execute scripts/lyria.py with uv run:

# Generate a full song with Lyria 3.5 (default model)
uv run scripts/lyria.py \
  -p "An energetic pop-rock song with driving drums and upbeat vocals" \
  -f "song.mp3" \
  -m "lyria-3.5" \
  --lyrics-file "lyrics.txt"

# Generate a 30-second chiptune arcade loop with Lyria 3 Clip
uv run scripts/lyria.py \
  -p "An energetic 8-bit chiptune arcade melody at 140 BPM in C major. Instrumental only." \
  -f "chiptune.mp3" \
  -m "clip"

# Multimodal Image-to-Music (compose music inspired by an image)
uv run scripts/lyria.py \
  -p "Ambient relaxing soundscape matching the mood of this landscape. Instrumental only." \
  -i "landscape.jpg" \
  -f "landscape_theme.mp3" \
  -m "lyria-3.5"

CLI Argument Reference

• -p, --prompt: Text prompt describing style, genre, instruments, BPM, and structure (required).
• -f, --filename: Output audio file path (default: music.mp3).
• -m, --model: lyria-3.5 / 3.5 (default), clip / lyria-3-clip-preview, or pro / lyria-3-pro-preview.
• -i, --input-image: Path to input image(s) for visual mood inspiration (up to 10).
• --format: Audio container format (mp3 or wav).
• --lyrics-file: Optional path to write generated lyric transcription / structure text.
• --api: interactions (default) or models.

2. Dual SDK Integration Patterns

Interactions API (client.interactions.create) — Recommended

import base64
from google import genai

client = genai.Client()

# Generate full-length song with timestamp structure
prompt = """
An atmospheric lo-fi beat in D Minor at 80 BPM:
[0:00 - 0:15] Intro: Dusty vinyl crackle and mellow electric piano chords.
[0:15 - 0:45] Verse: Warm boom-bap drum groove enters with subtle bassline.
[0:45 - 1:15] Chorus: Full melodic progression with saxophone accents.
[1:15 - 1:30] Outro: Slow fade out with piano alone.
"""

interaction = client.interactions.create(
    model="lyria-3.5",
    input=prompt,
)

if interaction.output_audio:
    with open("lofi_track.mp3", "wb") as f:
        f.write(base64.b64decode(interaction.output_audio.data))

if interaction.output_text:
    print("Generated Lyrics / Structure:\n", interaction.output_text)

Multimodal Image-to-Audio (Python)

import base64
from google import genai

client = genai.Client()

with open("art_concept.png", "rb") as f:
    img_b64 = base64.b64encode(f.read()).decode("utf-8")

interaction = client.interactions.create(
    model="lyria-3.5",
    input=[
        {"type": "image", "data": img_b64, "mime_type": "image/png"},
        {"type": "text", "text": "Compose an ethereal cyberpunk synth soundtrack matching this neon cityscape. Instrumental only."},
    ],
)

if interaction.output_audio:
    with open("cyberpunk_theme.mp3", "wb") as f:
        f.write(base64.b64decode(interaction.output_audio.data))

Prompting Best Practices

1. Genre & Subgenre: Be specific (e.g. synthwave, delta blues, chamber pop, lo-fi hip-hop).
2. Instrumentation: Name specific instruments (Fender Rhodes, TR-808 drums, nylon-string guitar, cello).
3. Tempo & Key: Include BPM and musical key (120 BPM, in A Minor).
4. Vocal Control: For instrumental tracks, explicitly specify "Instrumental only, no vocals". For vocals, provide bracketed lyrics ([Verse], [Chorus]).
5. Timestamp Arrangements: For Pro models, use bracketed timestamps ([0:00 - 0:20]) to direct transitions and builds.