Lyria 3 API: Pricing, Models, Examples and Full-Song Generation

Lyria 3 is Google's music-generation family on the Gemini API. Google documents text-to-music, image-conditioned generation, generated lyrics and song structure, and MP3 output. The current music-generation guide describes 44.1 kHz stereo audio. The Clip model card currently says 48 kHz stereo. Confirm the sample-rate figure against the docs you integrate against.

Do not describe Clip or Pro as generally available. Both IDs still use the preview suffix. Google's pricing page now lists Lyria 3 as a legacy family next to Lyria 3.5 (lyria-3.5), which the current Interactions guide uses for full-song examples.

Google currently documents both the Interactions API (interactions.create) and a legacy generateContent path. New examples on the music-generation guide use Interactions.

These are Google paid-tier list prices, per request/song, from the Gemini API pricing page. They are not OfflineCreator Studio credit prices. Do not convert Pro into a per-second rate: the unit is one song, and generated duration can vary.

Source: ai.google.dev/gemini-api/docs/pricing. Status notes: Clip and Pro remain preview IDs; Google currently calls Lyria 3 a legacy family. Lyria 3.5 is listed separately at $0.08 per full song.

Use Clip when iterating on mood, instruments, or prompt direction. Use Pro when you need a complete song structure. Google states that repeated calls can produce different results, so do not treat either model as deterministic.

These examples follow Google's current Interactions SDK pattern. They call Gemini directly. They are not OfflineCreator Studio request syntax.

Python

import base64
from google import genai

client = genai.Client()

interaction = client.interactions.create(
    model="lyria-3-clip-preview",
    input="Create a 30-second cinematic synth track with no vocals.",
)

generated_audio = interaction.output_audio
if generated_audio:
    with open("lyria-clip.mp3", "wb") as audio_file:
        audio_file.write(base64.b64decode(generated_audio.data))

lyrics = interaction.output_text
if lyrics:
    print(lyrics)

JavaScript

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

const client = new GoogleGenAI({});

const interaction = await client.interactions.create({
  model: "lyria-3-clip-preview",
  input: "Create a 30-second cinematic synth track with no vocals.",
});

if (interaction.output_audio) {
  fs.writeFileSync(
    "lyria-clip.mp3",
    Buffer.from(interaction.output_audio.data, "base64"),
  );
}

The legacy generateContent music guide still shows models.generateContent against the same Clip and Pro IDs. Verify the method your SDK version actually exposes before you ship.

Google documents Pro for full-length songs lasting a couple of minutes. Prompt wording can influence duration and arrangement. You can name sections such as verse, chorus, and bridge, and you can supply original lyrics. Exact second length is not guaranteed.

Google's current Interactions examples now pass lyria-3.5 for that full-song path. The Pro preview ID remains on the pricing table. If you need the current Google example path, start from lyria-3.5 and keep the Pro ID only when you have confirmed it still answers.

Create an energetic synth-pop song at 118 BPM.
Structure: intro, verse, chorus, verse, chorus, bridge, final chorus.
Use bright analog synths, punchy drums and original lyrics about exploring a new city at night.

Google documents image-conditioned generation. The current music-generation guide allows up to 10 images beside the text prompt for Lyria 3.5. The legacy generateContent guide documents the same image-count pattern for Lyria 3.

Typical uses are a soundtrack from concept art, music from a product mood board, scoring for a game scene, or adapting a visual campaign into audio. The model composes music inspired by the images; it does not attach a licensed soundtrack from the source picture.

• Clip and Pro remain preview IDs; Google now also lists Lyria 3 as a legacy family.
• Safety filters apply. Prompts that request specific artist voices or copyrighted lyrics may be blocked.
• Generated audio includes a SynthID audio watermark.
• Multi-turn iterative editing is not currently supported.
• Clip duration is fixed at 30 seconds.
• Pro duration is not guaranteed to an exact second.
• Results are non-deterministic even with the same prompt.
• Google's pricing table marks the free tier as not available.

No. The live OfflineCreator Studio catalog and provider routes inspected for this page do not include lyria-3-clip-preview, lyria-3-pro-preview, or lyria-3.5. Studio does not currently route Lyria. Do not send a Studio generation, MCP, or CLI request expecting a Lyria model ID.

Studio today publishes image, video, editing, and 3D models with option-aware credit quotes. Some video models generate native soundtrack audio as part of a video job. That is not a standalone music API and is not a Lyria route.

• Live Studio model catalog
• Studio credit pricing
• AI generation router
• Unified Studio generation API
• MCP and CLI
• Studio video generator

Only callable APIs with public documentation retrieved for this page are listed. Unverified prices are omitted.