You keep seeing "Lyria" attached to three different things — Lyria 3, Lyria 3 Pro, and Lyria RealTime — and Google's own pages don't make it obvious which one you actually want. Worse, most articles use the names interchangeably, so you end up expecting a full three-minute pop song and get a knob-twiddling live jam engine instead (or the reverse).
They are not the same tool, and the difference is not subtle. Lyria RealTime is a genuinely different kind of model: instead of generating a finished track you download, it streams music continuously while you steer it in real time. That makes it powerful for some jobs and completely wrong for others.
This guide explains what Lyria RealTime is, what it can and can't do, and — the part most explainers skip — a straight decision table for Lyria RealTime vs Lyria 3 vs Lyria 3 Pro, so you stop guessing which one your project needs. It's based on Google DeepMind's model pages, the Gemini API documentation, and Google's own launch coverage.
What Is Lyria RealTime?
Lyria RealTime is Google DeepMind's real-time, interactive music generation model. Rather than returning a single finished clip, it produces a continuous stream of instrumental music that you shape on the fly — changing the genre, mood, or instrumentation mid-stream and hearing the result within a couple of seconds.
Google opened it up through an API at Google I/O 2025 (May 20, 2025), and it powers the experimental MusicFX DJ app — the app that lets you blend prompts like "lo-fi" and "funk bass" into an evolving live mix. Think of it less as a songwriter and more as a session player who never stops and instantly follows your direction.
A few facts straight from Google's documentation:
- Continuous stream, not a file. It generates ongoing audio you play live, not a fixed-length track you render once and download.
- High-fidelity output. The stream is 48 kHz stereo audio, the same fidelity as the rest of the Lyria family.
- Low latency. There's a maximum of roughly 2 seconds between when you change a control and when you hear the effect — fast enough to feel interactive.
- Prompt blending. You can combine multiple weighted text prompts at once to mix genres, instruments, and moods rather than picking a single description.
- Direct musical controls. Beyond text, you can steer key, tempo, density, and brightness as live parameters.
- Instrumental only. Lyria RealTime does not generate vocals — it's built for beds, backing, and soundscapes, not sung songs.
Where do you get it? It's available in the Gemini API and Google AI Studio as an experimental model. "Experimental" is worth taking seriously: behavior, availability, and limits can shift without much notice, so it's better suited to prototyping and interactive experiences than a locked-down production pipeline.
How Lyria RealTime Is Different From "Normal" AI Music Generation
Most AI music tools — including Lyria 3 and Lyria 3 Pro — follow a prompt-in, track-out pattern: you write a description, wait, and receive a completed clip. Lyria RealTime inverts that loop. The music is already playing, and your prompts and controls nudge a living stream in whatever direction you want.
That single design choice cascades into everything:
- You perform it, you don't render it. There's no "generate" button you press once. You start a session and conduct it.
- It responds, it doesn't restart. Change "add a driving four-on-the-floor beat" and the existing stream evolves toward it — it doesn't throw away what was playing and start over.
- It has no fixed length. A generated Lyria 3 Pro song is capped at three minutes; a RealTime stream runs as long as your session does.
- It trades finished structure for control. Because it's continuous and instrumental, you don't get a neatly arranged verse-chorus-bridge song with vocals — you get an atmosphere you drive.
That's why the use cases are different too. Google positions Lyria RealTime for AI-assisted songwriting (jam along and capture ideas), interactive soundscapes for installations, and games — anywhere the music needs to react to a person or a system in the moment. If you need a shareable, structured song with lyrics, that's a job for Lyria 3 Pro, not RealTime.
Lyria RealTime vs Lyria 3 vs Lyria 3 Pro: Which to Use When
This is the comparison that actually resolves the confusion. All three are part of Google's Lyria family, share the 48 kHz fidelity ceiling, and carry Google's SynthID watermarking approach — but they solve different problems.
| Lyria RealTime | Lyria 3 | Lyria 3 Pro | |
|---|---|---|---|
| Core mode | Continuous live stream you steer | Short generated clip | Full generated song |
| Output length | Unlimited (runs with your session) | Up to 30 seconds | Up to 3 minutes |
| Vocals & lyrics | No — instrumental only | Yes | Yes |
| Song structure | None (evolving texture) | Basic | Intro / verse / chorus / bridge |
| Live control | Key, tempo, density, brightness, prompt blend | Prompt only | Prompt + timestamps + structure |
| Interactivity | Real-time (~2s response) | One-shot | One-shot |
| Where to get it | Gemini API, AI Studio (experimental) | Gemini app, API, credit studios | Gemini app, Vids, API, Vertex AI, credit studios |
| Best for | Games, installations, live jamming, songwriting sketches | Quick sketches, social clips, testing an idea | Finished songs, video soundtracks, releases |
Rule of thumb: if your music has to react to something happening right now, use Lyria RealTime. If you want a finished clip to test an idea, use Lyria 3. If you want a complete, structured song to ship, use Lyria 3 Pro.
For most creators reading this, the third row is the deciding one: RealTime never produces vocals, so anyone who wants a sung, structured track will land on Lyria 3 Pro. The credit-based Lyria 3 Pro music generator on this site runs exactly that model — full three-minute songs with vocals — without a Google Cloud project or a Gemini subscription. RealTime, by contrast, is something you build against through Google's API when you're coding an interactive experience.
When Lyria RealTime Is the Right Tool (and When It Isn't)
Reach for Lyria RealTime when:
- You're building a game or app where the soundtrack should shift with the player's state (calm exploration, tense combat, victory).
- You're creating an interactive installation or live visual where music needs to respond to sensors, crowds, or a performer.
- You want to jam and explore — blending prompts and pushing tempo/brightness to discover a direction before you commit it to a produced song.
- You specifically want instrumental, evolving beds rather than a fixed arrangement.
Skip it and use Lyria 3 Pro when:
- You need vocals and lyrics. RealTime doesn't sing.
- You need a shareable, downloadable song with a clear beginning, middle, and end.
- You need repeatable output — the same prompt yielding the same deliverable — for a client or a release.
- You want structural precision like a guaranteed bridge at a set timestamp.
A quick honesty check: because RealTime is experimental and API-only, it's not a "sign up and click generate" web tool for non-developers. If you don't want to write code against the Gemini API, it's likely not your path — and a credit studio running Lyria 3 Pro will get you a finished track far faster. You can compare the model tiers in more depth on our Lyria 3 overview, and the flagship complete Lyria 3 Pro guide walks through prompting full songs end to end.
How to Try Lyria RealTime (Low-Friction Path)
You don't need to build anything to hear what real-time generation feels like. The fastest verification step is to open Google's experimental MusicFX DJ app, which is powered by Lyria RealTime, and blend a couple of prompts — for example a "lo-fi hip hop" bed with "warm Rhodes piano," then nudge the tempo and density controls and listen to how quickly the stream reacts. That two-minute experiment tells you more about whether RealTime fits your project than any spec sheet.
If you're a developer, the model lives in Google AI Studio and the Gemini API under its experimental real-time music endpoints — Google's real-time music generation documentation is the place to start, and it covers the session model, prompt weighting, and the live controls.
If, after trying it, what you actually wanted was a finished song with vocals, that's the signal to switch tools. Write one sentence describing the track you want and generate a full song with Lyria 3 Pro instead — you'll have a structured, downloadable result in minutes, and you can check exact credit costs on the pricing page.
FAQ
What is Lyria RealTime in simple terms? It's Google DeepMind's interactive music model that streams continuous instrumental audio you can steer live — changing genre, mood, key, tempo, density, and brightness while the music keeps playing, with about a two-second response time.
Is Lyria RealTime the same as Lyria 3? No. Lyria 3 generates short finished clips (up to 30 seconds) and can include vocals; Lyria RealTime produces an endless, instrumental-only stream you control in real time. They're different models for different jobs.
Does Lyria RealTime have an API? Yes. It's available as an experimental model through the Gemini API and Google AI Studio, and it powers Google's MusicFX DJ app.
Can Lyria RealTime make songs with vocals? No. Lyria RealTime is instrumental only. For AI-generated vocals and lyrics in a structured song, you need Lyria 3 Pro.
What's the latency of Lyria RealTime? There's a maximum of roughly two seconds between changing a control or prompt and hearing the effect in the stream — low enough to feel interactive for live and game use.
Should I use Lyria RealTime or Lyria 3 Pro? Use Lyria RealTime for interactive, reactive, instrumental music (games, installations, live jamming). Use Lyria 3 Pro for finished, structured songs with vocals that you want to download and share.
Bottom Line
Lyria RealTime, Lyria 3, and Lyria 3 Pro sound like versions of one thing, but they're three answers to three different questions. RealTime is a live instrument — a continuous 48 kHz stream you conduct through key, tempo, density, brightness, and blended prompts, ideal for games, installations, and songwriting exploration, and available (experimentally) through the Gemini API. It doesn't do vocals, structure, or downloadable songs.
If that's what you were picturing when you searched "Lyria," you actually want Lyria 3 Pro. The quickest way to settle it is to hear both: try MusicFX DJ for the real-time feel, and run one prompt through the Lyria 3 Pro generator for a finished, structured song. Pick the tool that matches the job, not the one with the most familiar name.
Sources
Primary sources used for the factual claims in this article. Note: Lyria RealTime is an experimental model — capabilities and availability may change after publication (July 2026).
- Google DeepMind: Lyria RealTime — real-time model overview, controls, use cases
- Google DeepMind: Lyria model page — Lyria model family and capabilities
- Google AI for Developers: Lyria RealTime (experimental) — API availability, experimental status
- Google AI for Developers: Real-time music generation — session model, live controls, prompt blending
- Magenta: Introducing the Lyria RealTime API — 48 kHz streaming, latency, MusicFX DJ
- TechCrunch: Google brings Lyria RealTime to its API — I/O 2025 launch coverage
