Generate music with Lyria 3.5

Lyria 3.5 is Google's family of music generation models, available through the Gemini API. With Lyria 3.5, you can generate high-quality, 44.1 kHz stereo audio from text prompts or from images. These models deliver structural coherence, including vocals, timed lyrics, and full instrumental arrangements.

The Lyria family includes models:

Model Model ID Best for Duration Output
Lyria 3 Clip lyria-3-clip-preview Short clips, loops, previews 30 seconds MP3
Lyria 3.5 lyria-3.5 Full-length songs with verses, choruses, bridges A couple of minutes (controllable using prompt) MP3

Both models can be used using the new Interactions API, supporting multimodal inputs (text and images), and produce 44.1 kHz high-fidelity stereo audio.

Generate a music clip

The Lyria 3 Clip model always generates a 30-second clip. To generate a clip, call the interactions.create method with a text prompt. The response always includes the generated lyrics and song structure alongside the audio in the steps schema.

Python

import base64
from google import genai

client = genai.Client()

interaction = client.interactions.create(
    model="lyria-3-clip-preview",
    input="A short instrumental acoustic guitar piece.",
)

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

lyrics = interaction.output_text
if lyrics:
    print(f"Lyrics:\n{lyrics}")

JavaScript

import { GoogleGenAI } from '@google/genai';
import * as fs from 'fs';

const client = new GoogleGenAI({});

const interaction = await client.interactions.create({
    model: 'lyria-3-clip-preview',
    input: 'A short instrumental acoustic guitar piece.',
});

const generatedAudio = interaction.output_audio;
if (generatedAudio) {
  fs.writeFileSync('music.mp3', Buffer.from(generatedAudio.data, 'base64'));
}

const lyrics = interaction.output_text;
if (lyrics) {
  console.log(`Lyrics:\n${lyrics}`);
}

Java

import com.google.genai.Client;
import com.google.genai.gaos.models.interactions.CreateModelInteraction;
import com.google.genai.gaos.models.interactions.Interaction;
import com.google.genai.gaos.models.interactions.InteractionsInput;
import com.google.genai.gaos.models.interactions.Model;
import com.google.genai.gaos.models.operations.CreateInteractionRequestBody;
import java.nio.file.Files;
import java.nio.file.Paths;
import java.util.Base64;

Client client = new Client();

CreateModelInteraction params =
    CreateModelInteraction.builder()
        .model(Model.of("lyria-3-clip-preview"))
        .input(InteractionsInput.of("A short instrumental acoustic guitar piece."))
        .build();

Interaction interaction =
    client.interactions.create(CreateInteractionRequestBody.of(params)).interaction().get();

if (interaction.outputAudio().isPresent() && interaction.outputAudio().get().data().isPresent()) {
  byte[] audioBytes = Base64.getDecoder().decode(interaction.outputAudio().get().data().get());
  Files.write(Paths.get("music.mp3"), audioBytes);
}

interaction.outputText().ifPresent(lyrics -> System.out.println("Lyrics:\n" + lyrics));

REST

curl -X POST "https://generativelanguage.googleapis.com/v1beta/interactions" \
-H "Content-Type: application/json" \
-H "x-goog-api-key: $GEMINI_API_KEY" \
-d '{
    "model": "lyria-3-clip-preview",
    "input": "A short instrumental acoustic guitar piece."
}'

You can retrieve generated music data by using the interaction.output_audio property, which returns the last generated audio block. You can also retrieve the song's lyrics and structure by using the interaction.output_text property. For details on convenience properties, see the Interactions overview.

Generate a full-length song

Use the lyria-3.5 model to generate full-length songs that last a couple of minutes. The Pro model understands musical structure and can create compositions with distinct verses, choruses, and bridges. You can influence the duration by specifying it in your prompt (e.g., "create a 2-minute song") or by using timestamps to define the structure.

Python

interaction = client.interactions.create(
    model="lyria-3.5",
    input="An epic cinematic orchestral piece about a journey home. Starts with a solo piano intro, builds through sweeping strings, and climaxes with a massive wall of sound.",
)

JavaScript

const interaction = await client.interactions.create({
    model: 'lyria-3.5',
    input: 'A beautiful piano melody.',
});

Java

import com.google.genai.Client;
import com.google.genai.gaos.models.interactions.CreateModelInteraction;
import com.google.genai.gaos.models.interactions.Interaction;
import com.google.genai.gaos.models.interactions.InteractionsInput;
import com.google.genai.gaos.models.interactions.Model;
import com.google.genai.gaos.models.operations.CreateInteractionRequestBody;

Client client = new Client();

CreateModelInteraction params =
    CreateModelInteraction.builder()
        .model(Model.of("lyria-3.5"))
        .input(
            InteractionsInput.of(
                "An epic cinematic orchestral piece about a journey home. Starts with a solo piano intro, builds through sweeping strings, and climaxes with a massive wall of sound."))
        .build();

Interaction interaction =
    client.interactions.create(CreateInteractionRequestBody.of(params)).interaction().get();

REST

curl -X POST "https://generativelanguage.googleapis.com/v1beta/interactions" \
-H "Content-Type: application/json" \
-H "x-goog-api-key: $GEMINI_API_KEY" \
-d '{
    "model": "lyria-3.5",
    "input": "A beautiful piano melody."
}'

Select output format

By default, the Lyria 3.5 models generate audio in MP3 format. For Lyria 3.5, you can also request the output in WAV format by setting the response_format.

Python

interaction = client.interactions.create(
    model="lyria-3.5",
    input="A beautiful piano melody.",
    response_format={"type": "audio"},
)

JavaScript

const interaction = await client.interactions.create({
    model: 'lyria-3.5',
    input: 'A beautiful piano melody.',
    response_format: {
        type: 'audio',
    },
});

Java

import com.google.genai.Client;
import com.google.genai.gaos.models.interactions.AudioResponseFormat;
import com.google.genai.gaos.models.interactions.CreateModelInteraction;
import com.google.genai.gaos.models.interactions.CreateModelInteractionResponseFormat;
import com.google.genai.gaos.models.interactions.Interaction;
import com.google.genai.gaos.models.interactions.InteractionsInput;
import com.google.genai.gaos.models.interactions.Model;
import com.google.genai.gaos.models.interactions.ResponseFormat;
import com.google.genai.gaos.models.operations.CreateInteractionRequestBody;

Client client = new Client();

CreateModelInteraction params =
    CreateModelInteraction.builder()
        .model(Model.of("lyria-3.5"))
        .input(InteractionsInput.of("A beautiful piano melody."))
        .responseFormat(
            CreateModelInteractionResponseFormat.of(
                ResponseFormat.of(AudioResponseFormat.builder().build())))
        .build();

Interaction interaction =
    client.interactions.create(CreateInteractionRequestBody.of(params)).interaction().get();

REST

curl -X POST "https://generativelanguage.googleapis.com/v1beta/interactions" \
  -H "x-goog-api-key: $GEMINI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "lyria-3.5",
    "input": "A beautiful piano melody.",
    "response_format": {
        "type": "audio"
    }
  }'

Parse the response

The response from Lyria 3.5 contains multiple content blocks within the steps schema. Interactions return a sequence of steps, where model_output steps contain the generated content. Text content blocks contain the generated lyrics or a JSON description of the song structure. Content blocks with audio type contain the base64 encoded audio data.

Python

lyrics = []
audio_data = None

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

lyrics = interaction.output_text
if lyrics:
    print(f"Lyrics:\n{lyrics}")

JavaScript

const lyrics = [];
let audioData = null;

const generatedAudio = interaction.output_audio;
if (generatedAudio) {
    fs.writeFileSync("output.mp3", Buffer.from(generatedAudio.data, 'base64'));
}

const lyrics = interaction.output_text;
if (lyrics) {
    console.log("Lyrics:\n" + lyrics);
}

Java

import com.google.genai.Client;
import com.google.genai.gaos.models.interactions.CreateModelInteraction;
import com.google.genai.gaos.models.interactions.Interaction;
import com.google.genai.gaos.models.interactions.InteractionsInput;
import com.google.genai.gaos.models.interactions.Model;
import com.google.genai.gaos.models.operations.CreateInteractionRequestBody;
import java.nio.file.Files;
import java.nio.file.Paths;
import java.util.Base64;

Client client = new Client();

CreateModelInteraction params =
    CreateModelInteraction.builder()
        .model(Model.of("lyria-3.5"))
        .input(InteractionsInput.of("A song about a starry night."))
        .build();

Interaction interaction =
    client.interactions.create(CreateInteractionRequestBody.of(params)).interaction().get();

if (interaction.outputAudio().isPresent() && interaction.outputAudio().get().data().isPresent()) {
  byte[] audioBytes = Base64.getDecoder().decode(interaction.outputAudio().get().data().get());
  Files.write(Paths.get("output.mp3"), audioBytes);
}

if (interaction.outputText().isPresent()) {
  System.out.println("Lyrics:\n" + interaction.outputText().get());
}

REST

# The output from the REST API is a JSON object containing base64 encoded data.
# You can extract the text or the audio data using a tool like jq.
# To extract the audio and save it to a file:
curl ... | jq -r '.steps[] | select(.type=="model_output") | .content[] | select(.type=="audio") | .data' | base64 -d > output.mp3

Interleaved lyrics and music

Because the output from Lyria 3.5 is complex—containing separate steps and blocks for generated lyrics (text) and the song itself (audio)—convenience properties offer a fast and recommended shortcut.

However, if you want full, programmatic control over the raw timeline of steps returned by the server (such as logging individual content blocks as they are received), you can manually iterate over steps instead:

Python

lyrics = []
audio_data = None

for step in interaction.steps:
    if step.type == "model_output":
        for content_block in step.content:
            if content_block.type == "audio":
                audio_data = base64.b64decode(content_block.data)
            elif content_block.type == "text":
                lyrics.append(content_block.text)

if lyrics