使用 Lyria 3 生成音樂

Lyria 3 是 Google 的音樂生成模型系列，可透過 Gemini API 使用。使用 Lyria 3，你可以根據文字提示詞或圖片生成高品質的 44.1 kHz 立體聲音訊。這些模型可提供結構一致的音樂，包括人聲、歌詞時間碼和完整樂器編曲。

Lyria 3 系列包含兩種模型：

型號 模型 ID 適用情境 時間長度 輸出
Lyria 3 Clip lyria-3-clip-preview 短片、循環播放、預覽 30 秒 MP3
Lyria 3 Pro lyria-3-pro-preview 包含主歌、副歌和橋段的完整歌曲 幾分鐘 (可透過提示控制) MP3

這兩款模型都可透過標準 generateContent 方法和新的 Interactions API 使用，支援多模態輸入 (文字和圖片)，並產生 44.1 kHz 高傳真立體聲音訊。

生成音樂短片

Lyria 3 Clip 模型一律會生成 30 秒片段。如要生成短片，請使用文字提示呼叫 generateContent 方法。回應一律會包含生成的歌詞和歌曲結構，以及音訊。

Python

from google import genai

client = genai.Client()

response = client.models.generate_content(
    model="lyria-3-clip-preview",
    contents="Create a 30-second cheerful acoustic folk song with "
             "guitar and harmonica.",
)

# Parse the response
for part in response.parts:
    if part.text is not None:
        print(part.text)
    elif part.inline_data is not None:
        with open("clip.mp3", "wb") as f:
            f.write(part.inline_data.data)
        print("Audio saved to clip.mp3")

JavaScript

import { GoogleGenAI } from "@google/genai";
import * as fs from "node:fs";

const ai = new GoogleGenAI({});

async function main() {
  const response = await ai.models.generateContent({
    model: "lyria-3-clip-preview",
    contents: "Create a 30-second cheerful acoustic folk song with " +
              "guitar and harmonica.",

  });

  for (const part of response.candidates[0].content.parts) {
    if (part.text) {
      console.log(part.text);
    } else if (part.inlineData) {
      const buffer = Buffer.from(part.inlineData.data, "base64");
      fs.writeFileSync("clip.mp3", buffer);
      console.log("Audio saved to clip.mp3");
    }
  }
}

main();

Go

package main

import (
    "context"
    "fmt"
    "log"
    "os"

    "google.golang.org/genai"
)

func main() {
    ctx := context.Background()
    client, err := genai.NewClient(ctx, nil)
    if err != nil {
        log.Fatal(err)
    }

    result, err := client.Models.GenerateContent(
        ctx,
        "lyria-3-clip-preview",
        genai.Text("Create a 30-second cheerful acoustic folk song " +
                   "with guitar and harmonica."),
        nil,
    )
    if err != nil {
        log.Fatal(err)
    }

    for _, part := range result.Candidates[0].Content.Parts {
        if part.Text != "" {
            fmt.Println(part.Text)
        } else if part.InlineData != nil {
            err := os.WriteFile("clip.mp3", part.InlineData.Data, 0644)
            if err != nil {
                log.Fatal(err)
            }
            fmt.Println("Audio saved to clip.mp3")
        }
    }
}

Java

import com.google.genai.Client;
import com.google.genai.types.GenerateContentResponse;
import com.google.genai.types.Part;

import java.io.IOException;
import java.nio.file.Files;
import java.nio.file.Paths;

public class GenerateMusicClip {
  public static void main(String[] args) throws IOException {

    try (Client client = new Client()) {
      GenerateContentResponse response = client.models.generateContent(
          "lyria-3-clip-preview",
          "Create a 30-second cheerful acoustic folk song with "
              + "guitar and harmonica.");

      for (Part part : response.parts()) {
        if (part.text().isPresent()) {
          System.out.println(part.text().get());
        } else if (part.inlineData().isPresent()) {
          var blob = part.inlineData().get();
          if (blob.data().isPresent()) {
            Files.write(Paths.get("clip.mp3"), blob.data().get());
            System.out.println("Audio saved to clip.mp3");
          }
        }
      }
    }
  }
}

REST

curl -s -X POST \
  "https://generativelanguage.googleapis.com/v1beta/models/lyria-3-clip-preview:generateContent" \
  -H "x-goog-api-key: $GEMINI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "contents": [{
      "parts": [
        {"text": "Create a 30-second cheerful acoustic folk song with guitar and harmonica."}
      ]
    }]
  }'

C#

using System.Threading.Tasks;
using Google.GenAI;
using Google.GenAI.Types;
using System.IO;

public class GenerateMusicClip {
  public static async Task main() {
    var client = new Client();
    var response = await client.Models.GenerateContentAsync(
      model: "lyria-3-clip-preview",
      contents: "Create a 30-second cheerful acoustic folk song with guitar and harmonica."
    );

    foreach (var part in response.Candidates[0].Content.Parts) {
      if (part.Text != null) {
        Console.WriteLine(part.Text);
      } else if (part.InlineData != null) {
        await File.WriteAllBytesAsync("clip.mp3", part.InlineData.Data);
        Console.WriteLine("Audio saved to clip.mp3");
      }
    }
  }
}

生成完整歌曲

使用 lyria-3-pro-preview 模型生成幾分鐘的完整歌曲。Pro 版模型可瞭解音樂結構，並創作具有不同主歌、副歌和橋段的樂曲。如要影響時長，可以在提示中指定 (例如「創作 2 分鐘的歌曲」)，或使用時間戳記定義結構。

Python

response = client.models.generate_content(
    model="lyria-3-pro-preview",
    contents="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 response = await ai.models.generateContent({
  model: "lyria-3-pro-preview",
  contents: "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.",

});

Go

result, err := client.Models.GenerateContent(
    ctx,
    "lyria-3-pro-preview",
    genai.Text("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."),
    nil,
)

Java

GenerateContentResponse response = client.models.generateContent(
    "lyria-3-pro-preview",
    "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.");

REST

curl -s -X POST \
  "https://generativelanguage.googleapis.com/v1beta/models/lyria-3-pro-preview:generateContent" \
  -H "x-goog-api-key: $GEMINI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "contents": [{
      "parts": [
        {"text": "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."}
      ]
    }]
  }'

C#

var response = await client.Models.GenerateContentAsync(
  model: "lyria-3-pro-preview",
  contents: "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."
);

選取輸出格式

根據預設，Lyria 3 模型會以 MP3 格式生成音訊。如果是 Lyria 3 Pro，您也可以在 generationConfig 中設定 response_mime_type，要求以 WAV 格式輸出。

Python

response = client.models.generate_content(
    model="lyria-3-pro-preview",
    contents="An atmospheric ambient track.",
    config=types.GenerateContentConfig(
        response_modalities=["AUDIO", "TEXT"],
        response_mime_type="audio/wav",
    ),
)

JavaScript

const response = await ai.models.generateContent({
  model: "lyria-3-pro-preview",
  contents: "An atmospheric ambient track.",
  config: {
    responseModalities: ["AUDIO", "TEXT"],
    responseMimeType: "audio/wav",
  },
});

Go

config := &genai.GenerateContentConfig{
    ResponseModalities: []string{"AUDIO", "TEXT"},
    ResponseMIMEType:   "audio/wav",
}

result, err := client.Models.GenerateContent(
    ctx,
    "lyria-3-pro-preview",
    genai.Text("An atmospheric ambient track."),
    config,
)

Java

GenerateContentConfig config = GenerateContentConfig.builder()
    .responseModalities("AUDIO", "TEXT")
    .responseMimeType("audio/wav")
    .build();

GenerateContentResponse response = client.models.generateContent(
    "lyria-3-pro-preview",
    "An atmospheric ambient track.",
    config);

C#

var config = new GenerateContentConfig {
  ResponseModalities = { "AUDIO", "TEXT" },
  ResponseMimeType = "audio/wav"
};

var response = await client.Models.GenerateContentAsync(
  model: "lyria-3-pro-preview",
  contents: "An atmospheric ambient track.",
  config: config
);

REST

curl -s -X POST \
  "https://generativelanguage.googleapis.com/v1beta/models/lyria-3-pro-preview:generateContent" \
  -H "x-goog-api-key: $GEMINI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "contents": [{
      "parts": [
        {"text": "An atmospheric ambient track."}
      ]
    }],
    "generationConfig": {
      "responseModalities": ["AUDIO", "TEXT"],
      "responseMimeType": "audio/wav"
    }
  }'

剖析回應

Lyria 3 的回覆包含多個部分，文字部分包含生成的歌詞或歌曲結構的 JSON 說明。含有音訊位元組的部分為 inline_data。

Python

lyrics = []
audio_data = None

for part in response.parts:
    if part.text is not None:
        lyrics.append(part.text)
    elif part.inline_data is not None:
        audio_data = part.inline_data.data

if lyrics:
    print("Lyrics:\n" + "\n".join(lyrics))

if audio_data:
    with open("output.mp3", "wb") as f:
        f.write(audio_data)

JavaScript

const lyrics = [];
let audioData = null;

for (const part of response.candidates[0].content.parts) {
  if (part.text) {
    lyrics.push(part.text);
  } else if (part.inlineData) {
    audioData = Buffer.from(part.inlineData.data, "base64");
  }
}

if (lyrics.length) {
  console.log("Lyrics:\n" + lyrics.join("\n"));
}

if (audioData) {
  fs.writeFileSync("output.mp3", audioData);
}

Go

var lyrics []string
var audioData []byte

for _, part := range result.Candidates[0].Content.Parts {
    if part.Text != "" {
        lyrics = append(lyrics, part.Text)
    } else if part.InlineData != nil {
        audioData = part.InlineData.Data
    }
}

if len(lyrics) > 0 {
    fmt.Println("Lyrics:\n" + strings.Join(lyrics, "\n"))
}

if audioData != nil {
    err := os.WriteFile("output.mp3", audioData, 0644)
    if err != nil {
        log.Fatal(err)
    }
}

Java

List<String> lyrics = new ArrayList<>();
byte[] audioData = null;

for (Part part : response.parts()) {
  if (part.text().isPresent()) {
    lyrics.add(part.text().get());
  } else if (part.inlineData().isPresent()) {
    audioData = part.inlineData().get().data().get();
  }
}

if (!lyrics.isEmpty()) {
  System.out.println("Lyrics:\n" + String.join("\n", lyrics));
}

if (audioData != null) {
  Files.write(Paths.get("output.mp3"), audioData);
}

C#

var lyrics = new List<string>();
byte[] audioData = null;

foreach (var part in response.Candidates[0].Content.Parts) {
  if (part.Text != null) {
    lyrics.Add(part.Text);
  } else if (part.InlineData != null) {
    audioData = part.InlineData.Data;
  }
}

if (lyrics.Count > 0) {
  Console.WriteLine("Lyrics:\n" + string.Join("\n", lyrics));
}

if (audioData != null) {
  await File.WriteAllBytesAsync("output.mp3", audioData);
}

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 '.candidates[0].content.parts[] | select(.inlineData) | .inlineData.data' | base64 -d > output.mp3

根據圖片生成音樂

Lyria 3 支援多模態輸入，除了文字提示詞，你最多還可提供 10 張圖片，模型會根據視覺內容創作音樂。

Python

from PIL import Image

image = Image.open("desert_sunset.jpg")

response = client.models.generate_content(
    model="lyria-3-pro-preview",
    contents=[
        "An atmospheric ambient track inspired by the mood and "
        "colors in this image.",
        image,
    ],
)

JavaScript

const imageData = fs.readFileSync("desert_sunset.jpg");
const base64Image = imageData.toString("base64");

const response = await ai.models.generateContent({
  model: "lyria-3-pro-preview",
  contents: [
    { text: "An atmospheric ambient track inspired by the mood " +
            "and colors in this image." },
    {
      inlineData: {
        mimeType: "image/jpeg",
        data: base64Image,
      },
    },
  ],

});

Go

imgData, err := os.ReadFile("desert_sunset.jpg")
if err != nil {
    log.Fatal(err)
}

parts := []*genai.Part{
    genai.NewPartFromText("An atmospheric ambient track inspired " +
        "by the mood and colors in this image."),
    &genai.Part{
        InlineData: &genai.Blob{
            MIMEType: "image/jpeg",
            Data:     imgData,
        },
    },
}

contents := []*genai.Content{
    genai.NewContentFromParts(parts, genai.RoleUser),
}

result, err := client.Models.GenerateContent(
    ctx,
    "lyria-3-pro-preview",
    contents,
    nil,
)

Java

GenerateContentResponse response = client.models.generateContent(
    "lyria-3-pro-preview",
    Content.fromParts(
        Part.fromText("An atmospheric ambient track inspired by "
            + "the mood and colors in this image."),
        Part.fromBytes(
            Files.readAllBytes(Path.of("desert_sunset.jpg")),
            "image/jpeg")));

REST

curl -s -X POST \
  "https://generativelanguage.googleapis.com/v1beta/models/lyria-3-pro-preview:generateContent" \
  -H "x-goog-api-key: $GEMINI_API_KEY" \
  -H 'Content-Type: application/json' \
  -d "{
    \"contents\": [{
      \"parts\":[
          {\"text\": \"An atmospheric ambient track inspired by the mood and colors in this image.\"},
          {
            \"inline_data\": {
              \"mime_type\":\"image/jpeg\",
              \"data\": \"<BASE64_IMAGE_DATA>\"
            }
          }
      ]
    }]
  }"

C#

var response = await client.Models.GenerateContentAsync(
  model: "lyria-3-pro-preview",
  contents: new List<Part> {
    Part.FromText("An atmospheric ambient track inspired by the mood and colors in this image."),
    Part.FromBytes(await File.ReadAllBytesAsync("desert_sunset.jpg"), "image/jpeg")
  }
);

提供自訂歌詞

你可以自行撰寫歌詞，並加入提示。使用 [Verse]、[Chorus] 和 [Bridge] 等區段標記，協助模型瞭解歌曲結構：

Python

prompt = """
Create a dreamy indie pop song with the following lyrics:

[Verse 1]
Walking through the neon glow,
city lights reflect below,
every shadow tells a story,
every corner, fading glory.

[Chorus]
We are the echoes in the night,
burning brighter than the light,
hold on tight, don't let me go,
we are the echoes down below.

[Verse 2]
Footsteps lost on empty streets,
rhythms sync to heartbeats,
whispers carried by the breeze,
dancing through the autumn leaves.
"""

response = client.models.generate_content(
    model="lyria-3-pro-preview",
    contents=prompt,
)

JavaScript

const prompt = `
Create a dreamy indie pop song with the following lyrics:

[Verse 1]
Walking through the neon glow,
city lights reflect below,
every shadow tells a story,
every corner, fading glory.

[Chorus]
We are the echoes in the night,
burning brighter than the light,
hold on tight, don't let me go,
we are the echoes down below.

[Verse 2]
Footsteps lost on empty streets,
rhythms sync to heartbeats,
whispers carried by the breeze,
dancing through the autumn leaves.
`;

const response = await ai.models.generateContent({
  model: "lyria-3-pro-preview",
  contents: prompt,

});

Go

prompt := `
Create a dreamy indie pop song with the following lyrics:

[Verse 1]
Walking through the neon glow,
city lights reflect below,
every shadow tells a story,
every corner, fading glory.

[Chorus]
We are the echoes in the night,
burning brighter than the light,
hold on tight, don't let me go,
we are the echoes down below.

[Verse 2]
Footsteps lost on empty streets,
rhythms sync to heartbeats,
whispers carried by the breeze,
dancing through the autumn leaves.
`

result, err := client.Models.GenerateContent(
    ctx,
    "lyria-3-pro-preview",
    genai.Text(prompt),
    nil,
)

Java

String prompt = """
    Create a dreamy indie pop song with the following lyrics:

    [Verse 1]
    Walking through the neon glow,
    city lights reflect below,
    every shadow tells a story,
    every corner, fading glory.

    [Chorus]
    We are the echoes in the night,
    burning brighter than the light,
    hold on tight, don't let me go,
    we are the echoes down below.

    [Verse 2]
    Footsteps lost on empty streets,
    rhythms sync to heartbeats,
    whispers carried by the breeze,
    dancing through the autumn leaves.
    """;

GenerateContentResponse response = client.models.generateContent(
    "lyria-3-pro-preview",
    prompt);

C#

var prompt = @"
Create a dreamy indie pop song with the following lyrics:

[Verse 1]
Walking through the neon glow,
city lights reflect below,
every shadow tells a story,
every corner, fading glory.

[Chorus]
We are the echoes in the night,
burning brighter than the light,
hold on tight, don't let me go,
we are the echoes down below.

[Verse 2]
Footsteps lost on empty streets,
rhythms sync to heartbeats,
whispers carried by the breeze,
dancing through the autumn leaves.
";

var response = await client.Models.GenerateContentAsync(
  model: "lyria-3-pro-preview",
  contents: prompt
);

REST

curl -s -X POST \
  "https://generativelanguage.googleapis.com/v1beta/models/lyria-3-pro-preview:generateContent" \
  -H "x-goog-api-key: $GEMINI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "contents": [{
      "parts": [
        {"text": "Create a dreamy indie pop song with the following lyrics: ..."}
      ]
    }]
  }'

控制時間和結構

你可以使用時間戳記，在歌曲的特定時間點指定要執行的動作。這項功能有助於控制樂器進入的時間、歌詞的傳送時間，以及歌曲的進展方式：

Python

prompt = """
[0:00 - 0:10] Intro: Begin with a soft lo-fi beat and muffled
              vinyl crackle.
[0:10 - 0:30] Verse 1: Add a warm Fender Rhodes piano melody
              and gentle vocals singing about a rainy morning.
[0:30 - 0:50] Chorus: Full band with upbeat drums and soaring
              synth leads. The lyrics are hopeful and uplifting.
[0:50 - 1:00] Outro: Fade out with the piano melody alone.
"""

response = client.models.generate_content(
    model="lyria-3-pro-preview",
    contents=prompt,
)

JavaScript

const prompt = `
[0:00 - 0:10] Intro: Begin with a soft lo-fi beat and muffled
              vinyl crackle.
[0:10 - 0:30] Verse 1: Add a warm Fender Rhodes piano melody
              and gentle vocals singing about a rainy morning.
[0:30 - 0:50] Chorus: Full band with upbeat drums and soaring
              synth leads. The lyrics are hopeful and uplifting.
[0:50 - 1:00] Outro: Fade out with the piano melody alone.
`;

const response = await ai.models.generateContent({
  model: "lyria-3-pro-preview",
  contents: prompt,

});

Go

prompt := `
[0:00 - 0:10] Intro: Begin with a soft lo-fi beat and muffled
              vinyl crackle.
[0:10 - 0:30] Verse 1: Add a warm Fender Rhodes piano melody
              and gentle vocals singing about a rainy morning.
[0:30 - 0:50] Chorus: Full band with upbeat drums and soaring
              synth leads. The lyrics are hopeful and uplifting.
[0:50 - 1:00] Outro: Fade out with the piano melody alone.
`

result, err := client.Models.GenerateContent(
    ctx,
    "lyria-3-pro-preview",
    genai.Text(prompt),
    nil,
)

Java

String prompt = """
    [0:00 - 0:10] Intro: Begin with a soft lo-fi beat and muffled
                  vinyl crackle.
    [0:10 - 0:30] Verse 1: Add a warm Fender Rhodes piano melody
                  and gentle vocals singing about a rainy morning.
    [0:30 - 0:50] Chorus: Full band with upbeat drums and soaring
                  synth leads. The lyrics are hopeful and uplifting.
    [0:50 - 1:00] Outro: Fade out with the piano melody alone.
    """;

GenerateContentResponse response = client.models.generateContent(
    "lyria-3-pro-preview",
    prompt);

C#

var prompt = @"
[0:00 - 0:10] Intro: Begin with a soft lo-fi beat and muffled
              vinyl crackle.
[0:10 - 0:30] Verse 1: Add a warm Fender Rhodes piano melody
              and gentle vocals singing about a rainy morning.
[0:30 - 0:50] Chorus: Full band with upbeat drums and soaring
              synth leads. The lyrics are hopeful and uplifting.
[0:50 - 1:00] Outro: Fade out with the piano melody alone.
";

var response = await client.Models.GenerateContentAsync(
  model: "lyria-3-pro-preview",
  contents: prompt
);

REST

curl -s -X POST \
  "https://generativelanguage.googleapis.com/v1beta/models/lyria-3-pro-preview:generateContent" \
  -H "x-goog-api-key: $GEMINI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "contents": [{
      "parts": [
        {"text": "[0:00 - 0:10] Intro: ..."}
      ]
    }]
  }'

生成配樂

如要生成背景音樂、遊戲配樂或任何不需要人聲的音樂，可以提示模型生成純音樂曲目：

Python

response = client.models.generate_content(
    model="lyria-3-clip-preview",
    contents="A bright chiptune melody in C Major, retro 8-bit "
             "video game style. Instrumental only, no vocals.",
)

JavaScript

const response = await ai.models.generateContent({
  model: "lyria-3-clip-preview",
  contents: "A bright chiptune melody in C Major, retro 8-bit " +
            "video game style. Instrumental only, no vocals.",

});

Go

result, err := client.Models.GenerateContent(
    ctx,
    "lyria-3-clip-preview",
    genai.Text("A bright chiptune melody in C Major, retro 8-bit " +
               "video game style. Instrumental only, no vocals."),
    nil,
)

Java

GenerateContentResponse response = client.models.generateContent(
    "lyria-3-clip-preview",
    "A bright chiptune melody in C Major, retro 8-bit "
        + "video game style. Instrumental only, no vocals.");

C#

var response = await client.Models.GenerateContentAsync(
  model: "lyria-3-clip-preview",
  contents: "A bright chiptune melody in C Major, retro 8-bit " +
            "video game style. Instrumental only, no vocals."
);

REST

curl -s -X POST \
  "https://generativelanguage.googleapis.com/v1beta/models/lyria-3-clip-preview:generateContent" \
  -H "x-goog-api-key: $GEMINI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "contents": [{
      "parts": [
        {"text": "A bright chiptune melody in C Major, retro 8-bit video game style. Instrumental only, no vocals."}
      ]
    }]
  }'

生成不同語言的音樂

Lyria 3 會根據提示的語言生成歌詞。如要生成法文歌詞的歌曲，請用法文撰寫提示。模型會根據語言調整語音風格和發音。

Python

response = client.models.generate_content(
    model="lyria-3-pro-preview",
    contents="Crée une chanson pop romantique en français sur un "
             "coucher de soleil à Paris. Utilise du piano et de "
             "la guitare acoustique.",
)

JavaScript

const response = await ai.models.generateContent({
  model: "lyria-3-pro-preview",
  contents: "Crée une chanson pop romantique en français sur un " +
            "coucher de soleil à Paris. Utilise du piano et de " +
            "la guitare acoustique.",

});

Go

result, err := client.Models.GenerateContent(
    ctx,
    "lyria-3-pro-preview",
    genai.Text("Crée une chanson pop romantique en français sur un " +
               "coucher de soleil à Paris. Utilise du piano et de " +
               "la guitare acoustique."),
    nil,
)

Java

GenerateContentResponse response = client.models.generateContent(
    "lyria-3-pro-preview",
    "Crée une chanson pop romantique en français sur un "
        + "coucher de soleil à Paris. Utilise du piano et de "
        + "la guitare acoustique.");

C#

var response = await client.Models.GenerateContentAsync(
  model: "lyria-3-pro-preview",
  contents: "Crée une chanson pop romantique en français sur un " +
            "coucher de soleil à Paris. Utilise du piano et de " +
            "la guitare acoustique."
);

REST

curl -s -X POST \
  "https://generativelanguage.googleapis.com/v1beta/models/lyria-3-pro-preview:generateContent" \
  -H "x-goog-api-key: $GEMINI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "contents": [{
      "parts": [
        {"text": "Crée une chanson pop romantique en français sur un coucher de soleil à Paris. Utilise du piano et de la guitare acoustique."}
      ]
    }]
  }'

模型智慧

Lyria 3 會分析提示程序，根據提示透過音樂結構 (前奏、主歌、副歌、橋段等) 推理。這項程序會在生成音訊前執行，確保結構一致性和音樂性。

互動 API

您可以使用 Interactions API 存取 Lyria 3 模型，這個 API 提供統一的介面，可與 Gemini 模型和代理程式互動。可簡化複雜多模態用途的狀態管理和長時間執行的工作。

Python

from google import genai

client = genai.Client()

interaction = client.interactions.create(
    model="lyria-3-pro-preview",
    input="A melancholic jazz fusion track in D minor, " +
          "featuring a smooth saxophone melody, walking bass line, " +
          "and complex drum rhythms.",
)

for output in interaction.outputs:
    if output.text:
        print(output.text)
    elif output.inline_data:
         with open("interaction_output.mp3", "wb") as f:
            f.write(output.inline_data.data)
         print("Audio saved to interaction_output.mp3")

JavaScript

import { GoogleGenAI } from '@google/genai';

const client = new GoogleGenAI({});

const interaction = await client.interactions.create({
  model: 'lyria-3-pro-preview',
  input: 'A melancholic jazz fusion track in D minor, ' +
         'featuring a smooth saxophone melody, walking bass line, ' +
         'and complex drum rhythms.',
});

for (const output of interaction.outputs) {
  if (output.text) {
    console.log(output.text);
  } else if (output.inlineData) {
    const buffer = Buffer.from(output.inlineData.data, 'base64');
    fs.writeFileSync('interaction_output.mp3', buffer);
    console.log('Audio saved to interaction_output.mp3');
  }
}

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-pro-preview",
    "input": "A melancholic jazz fusion track in D minor, featuring a smooth saxophone melody, walking bass line, and complex drum rhythms."
}'

提示撰寫指南

提示詞可以很簡單，例如「一首關於可愛貓咪躲避水窪的民謠，女聲演唱，並加入雨聲」，也可以詳細且有結構，例如：

這首 1980 年代風格的合成器流行樂曲節奏強勁、合成器音色閃耀，副歌朗朗上口，令人振奮。歌曲應帶有復古未來感，讓人想起 80 年代的經典流行金曲，並以現代製作水準呈現。節奏應歡快且適合跳舞，每分鐘約 120 拍，並有清楚的主歌-副歌結構和令人難忘的樂器旋律。歌詞描述準備參加派對的心情。

簡單和複雜的提示都能產生良好的輸出內容。建議您嘗試這些訣竅，找出最適合自己的做法。

類型

在提示中加入想要的音樂類型，例如嘻哈、搖滾和饒舌。你可以指定多種曲風：

  • 融合金屬樂和饒舌
  • 結合死亡金屬和歌劇
  • 含有電子無人機元素的古典樂
  • 現代電子舞曲 (EDM) 混合歐洲流行音樂

你也可以加入年代：

  • 90 年代初期的嘻哈
  • 60 年代法國 ye-ye 流行樂
  • 80 年代的電子實驗
  • 2000 年代主流流行樂

如果提示詞要求生成特定類型或地區變體，例如「柏林鐵克諾」或「灣區 hyphy」，模型會嘗試捕捉該本質，但可能不一定能正確生成。

樂器

根據預設，Lyria 3 會使用該音樂類型常見的樂器和工具製作歌曲。不必提供具體的建議。

不過，除非你要求，否則舞曲不會包含薩克斯風。因此，如要生成薩克斯風獨奏，請輸入以下提示：

這首舞曲節奏強勁、合成器閃閃發光，副歌朗朗上口，橋段應加入薩克斯風獨奏。

提示可以包括特定樂器、樂器發出的聲音，以及樂器之間的互動方式。您可以運用這種組合營造特定情緒或質感：

  • 扭曲的低音線與乾淨俐落的 Hi-Hat 節奏對抗
  • 溫暖的類比合成器墊音在乾淨親切的木吉他下方膨脹
  • 多層模糊吉他音效打造的音牆，以及埋在其中的遠處人聲

歌曲結構

你可以在提示中列出歌曲的進展。使用箭頭或清單定義流程：

  • [Intro] -> [Verse 1] -> [Chorus] -> [Verse 2] -> [Chorus] -> [Bridge] -> [Outro]
  • 先以輕柔的鋼琴前奏開場，接著進入激昂的主歌，然後突然靜默，再爆發進入副歌。

你也可以指定這些區段之間的能量變化：

  • 在副歌前營造緊張感，然後在爆發力十足的副歌前歸於寂靜
  • 整首歌曲的音量逐漸變大，一次加入一種樂器，直到形成混亂的音牆
  • 在橋段後突然停止，接著是無伴奏合唱

你也可以提示要執行動作的確切時間：

  • 在 12 秒時建構至高點
  • 每隔 2 秒說一次「什麼」
  • 副歌從 22 秒開始

歌詞

系統預設會生成人聲和歌詞。你可以提供自己的歌詞、要求不要歌詞 (或純音樂)，或引導歌詞生成朝你想要的方向發展。

歌詞會以您輸入提示時使用的語言呈現。你也可以要求以其他語言撰寫歌詞，例如「用法文撰寫歌詞」。

使用自訂歌詞

如要讓模型使用你提供的歌詞，請在提示中加入歌詞，並加上「歌詞：」前置字元：

Lyrics:

[Intro]
Oooh, oooh

[Verse 1]
Let's go
Let's go
Go with the flow

[Chorus]
...

你可以在歌曲的各個部分加上前置字串，例如 [Intro]、[Verse 1]、[Pre-chorus]、[Chorus] 和 [Outro]。

如要重複某個字或一行，例如回音或和聲，請將其放在括號中：「Let's go (go)」。

提示模型撰寫歌詞

如要讓 Lyria 3 為你製作歌詞，建議在提示中加入歌詞內容的詳細資訊。否則模型必須從音樂提示推斷主題，結果可能不符合你的需求。

歌詞描述失去愛情的痛苦。這位歌手回憶起過去的戀情，以及隨之湧現的記憶。

如要重複播放副歌，請在提示中要求：

歌詞描述失去愛情的痛苦。這位歌手回憶起過去的戀情，以及隨之湧現的記憶。強而有力的副歌著重於克服痛苦並繼續前進。

Lyria 3 會根據你要求的音樂類型，自動調整歌詞結構，但你也可以在提示中重新強調這一點。例如：

EDM 歌曲，不斷重複相同的活力四射詞組。

你也可以提示加入非歌詞的聲音效果，例如：

  • 歌曲中不斷重複電影中的「我不敢相信！」片段
  • 在音樂即將進入高潮前，所有聲音都會停止，然後出現一個小小的聲音說「我不知道我在這裡做什麼」，接著音樂就會進入高潮。
  • 這首歌曲的開頭是關於 90 年代電影比現今電影更優秀的對話。接著，這首曲目會轉場至流行歌曲。

人聲

你可以提示歌詞的呈現方式，為獲得最佳效果，請詳細指定歌手的性別、音色和音域。

  • 女高音：音色清澈如水晶，靈活高亢。能以輕柔的氣音唱出高音。
  • 女中音：低音域豐富、溫暖且沙啞，帶有煙燻感，略帶聲帶震顫，充滿靈魂且共鳴感十足。
  • 男高音：明亮、高亢、充滿活力。音色年輕，略帶鼻音，高音爆發力十足，能穿透混音。
  • 男中音：深沉、醇厚，如天鵝絨般柔滑。渾厚胸腔共鳴聲，以輕柔的吟唱方式呈現。
  • 飽經風霜的搖滾歌手 (男聲)：嗓音沙啞粗獷，音色低沉，讓人想起 90 年代的垃圾搖滾。情緒強度過高。

其他提示參數

您也可以加入下列參數，進一步調整提示：

  • 調性/音階：指定音樂調性 (例如「G 大調」、「D 小調」)。
  • 情緒和氛圍：使用描述性形容詞 (例如「懷舊」、「激進」、「空靈」、「夢幻」)。
  • 長度：短片模型一律會生成 30 秒的短片。如果是 Pro 模型，請在提示中指定所需長度 (例如「創作 2 分鐘的歌曲」)，或使用時間戳記控制長度。

提示詞範例

以下列舉幾個有效的提示：

  • "A 30-second lofi hip hop beat with dusty vinyl crackle, mellow Rhodes piano chords, a slow boom-bap drum pattern at 85 BPM, and a jazzy upright bass line. Instrumental only."
  • "An upbeat, feel-good pop song in G major at 120 BPM with bright acoustic guitar strumming, claps, and warm vocal harmonies about a summer road trip."
  • "A dark, atmospheric trap beat at 140 BPM with heavy 808 bass, eerie synth pads, sharp hi-hats, and a haunting vocal sample. In D minor."

最佳做法

  • 先使用 Clip 進行疊代。使用速度較快的 lyria-3-clip-preview 模型測試提示，再使用 lyria-3-pro-preview 生成完整長度的內容。
  • 提供清楚明確的說明，模糊不清的提示會產生一般結果。提及樂器、BPM、調性、情境和結構，以獲得最佳輸出內容。
  • 使用章節標記。[Verse]、[Chorus]、[Bridge] 標記可為模型提供明確的結構。
  • 歌詞和指示分開。提供自訂歌詞時，請清楚區隔歌詞和音樂方向指示。

限制

  • 安全性：所有提示都會經過安全篩選器檢查。如果提示觸發篩選器，系統就會封鎖提示。包括要求特定藝人聲音的提示，或是生成受著作權保護的歌詞。
  • 浮水印：所有生成的音訊都會加上 SynthID 音訊浮水印，以利識別。這種浮水印人耳無法辨識，不會影響聆聽體驗。
  • 多輪編輯：生成音樂是單輪程序。 目前版本的 Lyria 3 不支援透過多個提示詞，反覆編輯或修正生成的片段。
  • 長度：片段模型一律會生成 30 秒的片段。Pro 模型會生成幾分鐘的歌曲，確切時長取決於提示。
  • 決定性：即使使用相同提示，每次呼叫的結果也可能不同。

後續步驟