Skip to content

Google Lyria

What it is

Lyria is Google's music-generation model family from DeepMind.

What problem it solves

It gives creators and product teams a model-native way to generate or assist with music creation.

Where it fits in the stack

AI & Knowledge / Generative Audio Model. It belongs in the media-generation layer of the AI landscape.

Typical use cases

  • Music generation experiments
  • Creative tooling and soundtrack prototyping
  • Exploring multimodal generative product ideas

Strengths

  • Backed by DeepMind's generative media research
  • Useful reference point in the music-generation landscape

Limitations

  • Access, productization, and workflow fit may be constrained
  • Music-generation tools often have rights and usage questions that must be reviewed carefully

When to use it

  • When evaluating music-generation capabilities in the Google ecosystem

When not to use it

  • When your use case needs a fully open or self-hosted audio stack

Getting started

Accessing Lyria

  1. Google Labs: Many Lyria-powered experiments (like Dream Track for YouTube Shorts) are rolled out via Google Labs.
  2. YouTube Creator Tools: If you are a recognized creator, check your YouTube Studio "Create" options for AI-assisted music generation features.
  3. Google AI Studio: Monitor the "Audio" or "Multimodal" model selections in Google AI Studio for experimental music generation endpoints as they become available.

Vertex AI (Enterprise)

To use Lyria via Google Cloud Vertex AI, you need a project with the Vertex AI API enabled.

# Install the Google Cloud AI Platform SDK
pip install google-cloud-aiplatform

import vertexai
from vertexai.generative_models import GenerativeModel

# Initialize Vertex AI
vertexai.init(project="your-project-id", location="us-central1")

# Note: Lyria access is typically gated via Model Garden
# This is a representative example of calling a multimodal audio model
model = GenerativeModel("lyria-002")

response = model.generate_content(
    "A 30-second lofi hip hop beat with mellow piano and slow boom-bap drums."
)

print(response.text)

Replicate (Community API)

If you prefer a simpler REST API, Lyria-3 is available via Replicate.

pip install replicate
export REPLICATE_API_TOKEN=your_token_here

import replicate

output = replicate.run(
    "google/lyria-3",
    input={
        "prompt": "A calm acoustic folk song with gentle guitar and soft strings."
    }
)
print(output.url)

CLI examples

You can trigger Lyria generations via the gcloud CLI if you have the appropriate permissions and model access.

# Print an access token for authentication
export AUTH_TOKEN=$(gcloud auth print-access-token)

# Send a raw prediction request to the Vertex AI endpoint
curl -X POST \
  -H "Authorization: Bearer $AUTH_TOKEN" \
  -H "Content-Type: application/json" \
  https://us-central1-aiplatform.googleapis.com/v1/projects/YOUR_PROJECT_ID/locations/us-central1/publishers/google/models/lyria-002:predict \
  -d '{
    "instances": [
      { "prompt": "An upbeat 120 BPM pop track with bright acoustic guitar." }
    ]
  }'

API examples

The Vertex AI REST API allows for fine-tuned control over generation parameters.

Requesting multiple samples

curl -X POST \
  -H "Authorization: Bearer $(gcloud auth print-access-token)" \
  -H "Content-Type: application/json" \
  https://us-central1-aiplatform.googleapis.com/v1/projects/YOUR_PROJECT_ID/locations/us-central1/publishers/google/models/lyria-002:predict \
  -d '{
    "instances": [{ "prompt": "Dark atmospheric trap beat" }],
    "parameters": { "sample_count": 2 }
  }'

Python SDK with specific parameters

from google.cloud import aiplatform

aiplatform.init(project="your-project-id")

endpoint = aiplatform.Endpoint("your-endpoint-id")
response = endpoint.predict(
    instances=[{"prompt": "Jazzy upright bass line with dusty vinyl crackle"}],
    parameters={"sample_count": 1}
)
print(response.predictions)

Sources / References

Contribution Metadata

  • Last reviewed: 2026-06-27
  • Confidence: high