Lyria 3 API Pricing: Lyria 3.5, Realtime and Google Access

Lyria 3 API pricing comes down to one number per request: $0.04 for a 30-second clip and $0.08 for a full song. There is no per-second meter, no token math for the audio, and no free tier on any of the paid models. Google also runs a streaming model called Lyria RealTime that has no published price, plus a consumer route through Flow Music and the Gemini app that skips the API entirely. This article lays out every rate, model ID and access path in one place, then runs the numbers for real projects so you can budget before you write a single line of code.

💡 Quick answer: Lyria 3.5 costs $0.08 per song in the Gemini API, the same as Lyria 3 Pro. The 30-second Clip model costs half of that at $0.04. Lyria RealTime is experimental and has no listed rate.

What Lyria 3 Costs Today

Google's Gemini API pricing page lists three Lyria models, and all three bill per generated output. Here they are next to the older Lyria 2, which still runs on Google Cloud.

None of the Gemini API models has a free tier, so a billing account has to be attached before the first request goes through. That is the biggest surprise for developers who are used to testing other Gemini models for nothing: with Lyria, even the first experiment is a paid call.

Per Song, Not Per Second

Prices are per generated output, so a 184-second track from Lyria 3 Pro costs the same as a shorter one. That changes how you should read the list. Convert each model to a cost per minute of finished audio and the cheap-looking Clip model turns out to be the most expensive way to buy music.

Use the Clip model for drafts, loops and quick previews. When the track is going to be heard by real people, pay for a full song model, because the extra $0.04 buys four to six times more audio.

The pricing page shows one rate per song and no separate line for reference images, lyrics or timestamps. Budget on the per-song number, then check your first invoice against it.

Lyria 3.5 in the Gemini API

Lyria 3.5 reached developers in two steps. Google launched it inside Flow Music on July 29, 2026, then opened it through the Gemini API, Google AI Studio and the Gemini app on September 3 and 4. The model ID is lyria-3.5, and Google documents it as the successor to Lyria 3 Pro.

Release Timeline

• May 20, 2025: Lyria 2 (lyria-002) arrives on Google Cloud with 30-second instrumental clips.
• March 25, 2026: Lyria 3 Clip and Lyria 3 Pro appear in preview.
• July 29, 2026: Lyria 3.5 launches in Flow Music.
• September 3 to 4, 2026: Lyria 3.5 lands in the Gemini API, AI Studio and the Gemini app.

What Changed From Pro

The price did not move. Lyria 3.5 costs $0.08 per song, exactly like Lyria 3 Pro, so the upgrade costs nothing extra. For the same money you get:

• Better music: richer melodic structure, closer lyric adherence and more expressive vocals.
• Two formats: MP3 by default and WAV on request, both at 44.1 kHz stereo.
• Image inputs: up to 10 reference images next to the text prompt.
• Structure control: section tags such as [Verse], [Chorus] and [Bridge], plus timestamps like [0:00 - 0:10] Intro:.
• Watermarking: every file carries a SynthID audio watermark.

Which one should you call? If your code already points at lyria-3-pro-preview, nothing breaks today. But both cost $0.08, Google positions 3.5 as the replacement, and preview model IDs are never a safe long-term bet, so new projects should start on lyria-3.5.

The limits matter as much as the features. Generation is single turn, so a track cannot be edited afterwards. You change the prompt and generate again. Requests for a specific artist's voice or for copyrighted lyrics are blocked, function calling and the Live API are not available, and the input limit is 131,072 tokens. Identical prompts return different songs, so treat every call as a sample and not a promise.

The call itself is short. This is the shape shown in the documentation:

from google import genai

client = genai.Client()
interaction = client.interactions.create(
    model="lyria-3.5",
    input="Warm indie folk, fingerpicked guitar, 92 BPM, female vocals. [Verse] [Chorus] [Verse] [Chorus]",
)

Copy Prompt

💡 Prompt tip: name the genre, the instruments, the tempo in BPM, the musical scale, the mood, and whether you want vocals. Keep the lyrics separate from the production notes so the model does not sing your instructions.

Lyria RealTime Pricing and Limits

Lyria RealTime is a different animal. Instead of one request and one finished file, you open a WebSocket and steer a music stream that keeps playing until you close the session.

How the Streaming Session Works

• Model name: models/lyria-realtime-exp, marked experimental and available on the v1beta API.
• Connection: a persistent, bidirectional, low-latency WebSocket.
• Audio: raw 16-bit PCM at 48 kHz stereo, always watermarked.
• Steering: weighted text prompts blend genres, instruments and moods on the fly.
• Controls: BPM from 60 to 200, density and brightness from 0.0 to 1.0, plus guidance, scale and mute options.
• Session commands: play, pause, stop and reset context. Changing the BPM or the scale requires a context reset.

What Google Has Not Priced

The Gemini API pricing page has no Lyria RealTime line. The documentation does not mention a rate either. That leaves developers with an experimental model, no published price and no promise that the current terms will last.

My advice: prototype with RealTime if an adaptive soundtrack is the whole point of your product, but do not build a paid plan around it until Google posts a rate. For anything that needs a predictable bill, use the per-song models.

Every Way to Get Google Access

There are four doors into Lyria, and only one of them involves writing code against Google directly.

Gemini API and AI Studio

This is the developer route. You create credentials in Google AI Studio, attach billing, and call lyria-3.5, lyria-3-pro-preview or lyria-3-clip-preview from Python, JavaScript, Java, Go or plain REST. Both the standard generateContent method and the newer Interactions API work. AI Studio doubles as a playground for testing prompts before you spend on code.

Flow Music and the Gemini App

Flow Music is where Lyria 3.5 launched, and it is free to use, with daily credits and two generations at a time on the free plan. Paid plans reported at launch start around $8 a month and scale up to $64. Commercial rights are expressly listed for the Google AI Plus, Pro and Ultra memberships, while the free plan does not publicly state them, so confirm the current terms before you monetize anything.

Google Cloud and Third-Party Gateways

Google Cloud hosts Lyria 2 at $0.06 per 30 seconds and the Lyria 3 preview models at the same $0.04 and $0.08 rates. Pick this door when you need a Google Cloud project, regional controls or enterprise billing. Gateways such as OpenRouter also list the Lyria 3 preview models, but check their rates directly since they can differ from Google's own.

The fourth door skips Google's billing altogether: a browser platform that hosts the models for you, described in the PicassoIA section below. Here is how the four routes line up:

Real Monthly Cost Examples

Raw prices are easy. Monthly bills are what matter. This table shows what volume does at the two public rates.

Indie Game Studio

A small studio scores 40 levels with 3 tracks each, so 120 tracks. Each track gets 5 Clip drafts to find a direction ($24 in total), then 2 full songs from Lyria 3.5 to land the final version ($19.20). The whole soundtrack costs $43.20.

Podcast Network

A network with 30 episodes a month wants an original music bed for each one. At 3 generations per episode on Lyria 3.5, that is 90 songs and $7.20 a month. About the price of a sandwich for the whole network.

Retry Overhead

The number that decides your real bill is the acceptance rate. You pay for every generation, including the ones you throw away. The effective cost of a usable song is the price divided by the share of tracks you keep.

Picture a consumer app with 5,000 users who each make 4 songs a month. That is 20,000 songs, or $1,600 if every result is accepted and $3,200 if half are rejected. Four habits keep that rate high:

• Draft with the $0.04 Clip model, then pay for the full song only once the direction is right.
• Use a fixed prompt template with genre, tempo, mood and structure tags, so results stay predictable.
• Store every accepted track and never regenerate the same prompt twice.
• Set a billing alert in your Google account and log each request next to a kept or rejected flag.

How to Use Lyria 3 on PicassoIA

If you want the Lyria sound without a billing account, a Google Cloud project or any code, PicassoIA runs the models in the browser. The AI music collection currently lists Lyria 3, Lyria 3 Pro and Lyria 2. Lyria 3.5 is not listed yet, so Lyria 3 Pro is the closest full-song option.

Lyria 3 or Lyria 3 Pro

Lyria 3 turns a text prompt into a 30-second clip in seconds, which makes it ideal for loops, jingles and social posts. Lyria 3 Pro builds full songs up to three minutes long with verses, choruses and bridges. Both accept up to 10 reference images, so a photo of a rainy street can set the mood of the track.

Five Steps to a Finished Track

1. Open the Lyria 3 Pro page.
2. Write a prompt with the genre, instruments, tempo, mood and vocal direction.
3. Add structure with tags like [Verse] and [Chorus], or lock moments with timestamps such as [0:00 - 0:30].
4. Optionally attach up to 10 reference images to steer the atmosphere.
5. Generate, listen, change one word in the prompt and generate again until it fits.

Here is a prompt pattern that works well:

[0:00 - 0:15] Intro: Soft lo-fi beat with vinyl crackle and distant rain.
[0:15 - 0:45] Verse: Warm Rhodes piano at 85 BPM, gentle bass, light percussion.
[0:45 - 1:15] Chorus: Full arrangement with lush pads and subtle strings.
Genre: Lo-fi chill hop. Mood: Cozy and reflective. Instrumental only.

Copy Prompt

Treat the browser as a sandbox. Test the structure and wording there, then paste the winning prompt into your API call, so your paid requests start from a prompt that already works.

💡 Want a second opinion? Run the same brief through MiniMax Music 2.6, ElevenLabs Music or Stable Audio 2.5 and compare the results side by side.

Add Voices Around Your Music

A soundtrack rarely works alone. Most projects that need a Lyria track also need a narrator, and sometimes a transcript of what was sung.

Voiceovers for Your Tracks

Gemini 3.1 Flash TTS offers 30 voices across more than 70 languages. ElevenLabs v3 and MiniMax Speech 2.8 HD suit studio-quality narration, and Inworld Realtime TTS 2 fits low-latency voice work. Generate the voice first, then write the music prompt around its pace.

Check Vocals With Transcription

Vocals are where AI music slips. A quick transcript shows whether the lyrics came out as written. Gemini 3 Pro, GPT 4o Transcribe and GPT 4o Mini Transcribe all turn audio into text. Compare the output with your lyric sheet before you publish.

Make Your First Track Today

Lyria 3 API pricing is easy to remember: $0.04 buys a clip, $0.08 buys a song, and RealTime waits for a price. The cheapest way to find out whether the sound fits your project is to listen before you build anything. Open Lyria 3 Pro on PicassoIA, type the track you imagine, and hear it in minutes. Then add a narrator with Gemini 3.1 Flash TTS, check the lyrics with Gemini 3 Pro, and keep experimenting with new prompts. The platform also builds photorealistic images and video for your album art and promos, so a whole release can come together in one place.