Make music with Google's Lyria models in AI Gateway

Your app can write songs now. AI Gateway serves Google’s Lyria music models on the Gemini Interactions API, so the access key and credits that already answer your chat requests can also commission a sea shanty.

So we commissioned one. We asked Lyria 3.5 for a shanty about a webhook that never gets a 200 OK. It came back as a three-minute song with an accordion, a lead singer who has clearly been on call too long, and a crew stomping through this chorus:

Then weigh the anchor, clear the bay,
We’ll retry till the break of day!

Two Lyria models run on ngrok-managed inference:

Use the clip model for game loops, video intros, and hold music, and the full model when your app needs an actual song. Google prices Lyria per request, not per token, so a 45-second jingle and a three-minute epic from Lyria 3.5 cost the same. AI Gateway charges that price to your ngrok credits, plus the standard processing fee, which came to $0.00017 on our shanty. Check Model Catalog for current pricing. With your own Google key attached to the gateway, Google bills you directly.

In the playground, pick Lyria 3.5 or Lyria 3 Clip Preview and describe a song. The reply comes back with a player you can play, seek, and Download, and the lyrics sit under Transcript. Stop is off while a song generates, because Google bills the song even if you cut the request short.

Get started

Create an ngrok account or sign in. Create an access key that allows Google, and confirm your account has credits. To use your own Google key instead, attach it to the gateway.

To hear the shanty before you write any code, send the prompt below in the playground with Lyria 3.5.

This example uses Google’s Gen AI SDK for TypeScript, because Lyria runs on Google’s native Interactions API. The Google Gen AI SDK guide covers the rest of the SDK setup. Install the SDK and a TypeScript runner:

npm install @google/genai && npm install --save-dev tsx

Save your ngrok access key in .env.local. Keep this file out of version control:

AI_GATEWAY_API_KEY=your_access_key_here

The example asks for the same shanty, prints its lyrics, and saves the song. Use an ES module project ("type": "module" in package.json) for top-level await. Save this example as index.ts:

1import { writeFile } from "node:fs/promises";2import { GoogleGenAI } from "@google/genai";3 4const client = new GoogleGenAI({⋯5	apiKey: process.env.AI_GATEWAY_API_KEY,6	httpOptions: {⋯7		baseUrl: "https://gateway.ngrok.ai",8		apiVersion: "v1beta",9	},10});11 12const interaction = await client.interactions.create({⋯13	model: "lyria-3.5",14	input:15		"A rowdy sea shanty about a webhook that keeps retrying because nobody ever sends back a 200 OK. Accordion, stomping boots, and a crew shouting the chorus.",16});17 18console.log(interaction.output_text);19 20const song = interaction.output_audio;21if (song?.data) {⋯22	await writeFile("shanty.mp3", Buffer.from(song.data, "base64"));23	console.log("Saved shanty.mp3");24}

Run the example from your project directory. A full song takes a little while, so give it about half a minute:

node --env-file=.env.local --import tsx index.ts

interaction.output_text holds the lyrics, and interaction.output_audio.data holds the song as base64 MP3. The lyrics keep Lyria’s song markup, such as [[A0]] and [:], so strip it before you show the words to a person. Here’s the start of one real run. Your crew sings a different shanty every time:

1[[A0]]2[:] Come all you brave coders and listen to me,3[:] Of a webhook that sailed on a boundless IP.4...5[:] (Retry!) (Retry!) (Give me that 200!)6Saved shanty.mp3

For a WAV file instead of MP3, add response_format: { type: "audio", mime_type: "audio/wav" } to the request. The price stays the same. For a 30-second loop, change model to "lyria-3-clip-preview".

Open Activity, find the request that wrote your shanty, and inspect its model, provider, and status. Usage shows 1 request, and Cost shows the song price plus the processing fee.