Google DeepMind just dropped Lyria 3 into the Gemini app recently, and it’s turning heads fast. The model can spin a full 30-second track, complete with vocals (which is quite impressive), lyrics, and instrumentation, from a single text prompt or photo. With rival Suno pulling in a $250 million funding round just months ago, the stakes couldn’t be higher. Google is not here to experiment quietly. Let’s cover the Lyria 3 release and show you how to use it on the Gemini app.

What Lyria 3 Actually Does

Lyria 3 is Google DeepMind’s most advanced generative music model yet, and calling it an incremental upgrade would be underselling it. The jump from Lyria 2 is meaningful in three specific areas: automatic lyric generation, improved creative control over style and tempo, and noticeably more realistic and musically complex output.

With previous versions, users had to supply their own lyrics or work around the model’s limitations. Lyria 3 handles that automatically, building words from the prompt itself. A user can type something like “a comical R&B slow jam about a sock finding its match,” and Gemini will produce a fully formed track in seconds, without any additional input required. That kind of friction reduction matters for mainstream adoption.

The model also works across modalities. Beyond text prompts, users can upload a photo or video and have Lyria 3 compose a matching soundtrack. It’s a genuinely clever multimodal feature, and one that broadens the tool’s appeal well beyond hobbyists. Each generated track also gets paired with custom cover art, courtesy of Nano Banana, Google’s own AI image generation model.

How It Compares to the Competition

The AI music space is crowded, but a few names keep coming up. Here’s how Lyria 3 stacks up against its closest rivals at launch:

♫	Google Lyria 3	Suno	Udio
Track length	30 seconds	Up to 4 minutes	Up to 3 minutes
Auto lyrics	Yes	Yes	Yes
Image-to-music	Yes	No	No
Integrated platform	Gemini / YouTube	Standalone app	Standalone app
AI watermarking	SynthID (imperceptible)	None confirmed	None confirmed
Free tier	Yes (usage limits)	Yes (10 songs/day)	Yes (limited)
Editing tools	Limited (beta)	Suno Studio DAW	Limited

The track length limitation is real, and Google has acknowledged it. Thirty seconds is enough for a Shorts backing track or a fun birthday message, but it won’t satisfy anyone hoping to produce a full song. Suno, by comparison, lets paid users generate tracks up to four minutes long and even offers a virtual DAW for post-generation editing. That’s a meaningful gap, and it’s one Google will need to close if it wants Lyria 3 to compete seriously in the creator economy.

Google Lyria 3 Key Features

Here’s what you get with Lyria 3 in Gemini right now:

• Multimodal input: Generate music from text descriptions, photos, or video uploads, giving creators flexible starting points.
• Auto-generated lyrics: No need to write your own words. The model builds lyrics directly from your prompt and handles the creative lift.
• Genre and style range: From pop and funk to Motown and ambient, Lyria 3 covers a wide tonal spectrum with real versatility.
• Custom cover art: Every track comes paired with AI-generated artwork produced by Nano Banana, making the output feel complete and shareable.
• SynthID watermarking: All tracks carry an imperceptible audio watermark that persists even through MP3 compression or analog re-recording, identifying the content as AI-generated.
• YouTube Dream Track integration: Lyria 3 now powers Dream Track for Shorts creators, expanding beyond the US to international markets for the first time.
• Multilingual support: Available in English, German, Spanish, French, Hindi, Japanese, Korean, and Portuguese at launch, with more languages in progress.
• Broad accessibility: Free for all Gemini users 18 and older, with higher usage caps for Google AI Plus, Pro, and Ultra subscribers.

How to Use Google Lyria 3 on Gemini

To use Lyria 3 on the Gemini website/app, select “Create music” from the tools ↓↓↓↓

Enter your prompt and make it as descriptive as possible (see the example below) ↓↓↓↓

Get your results ↓↓↓↓

The Safety and Copyright Picture

One of the louder conversations in AI music right now is around training data and artist rights. Google has not been immune to scrutiny here. Billboard reported back in January 2024 that Google had previously trained earlier AI music models on copyrighted recordings before seeking licensing agreements with rights holders. That history follows Lyria 3 to its launch, whether Google wants it to or not.

The company’s current position is that Lyria 3 is trained on music that YouTube and Google have the right to use under their terms of service, partner agreements, and applicable law. That’s a carefully worded statement, and the music industry knows it. Universal Music Group and YouTube did finalize a new licensing deal in October 2025, which UMG CEO Sir Lucian Grainge said included meaningful guardrails around generative AI content. That deal likely helped smooth the path to this launch.

On the output side, Google has implemented several protective measures worth knowing:

• If a prompt names a specific artist, Gemini treats it as stylistic inspiration only, not a directive to clone that artist’s voice or sound.
• Filters check generated outputs against existing content to reduce the risk of reproducing copyrighted material.
• SynthID watermarks are baked into every track at generation, providing a technical layer of attribution that survives heavy post-processing.
• Users can report content that may infringe on their rights or the rights of others directly within the app.

None of this is bulletproof, and Google has admitted as much. But the infrastructure is more robust than what most competitors have publicly committed to. The Lyria RealTime API, available for developers, adds another layer of transparency by enabling real-time steering of the generation process through something called WeightedPrompts, which gives technical users more auditable control over what the model is producing.

What the Music AI Sandbox Adds for Pros

Casual users get Gemini’s text-to-music interface. But Google also runs the Music AI Sandbox, a separate environment built specifically for working musicians and developers. It’s a different product with a different audience, and it reflects Google’s effort to position Lyria 3 as more than a toy.

Inside the Sandbox, users can do things that go well beyond basic prompting:

• Transform a simple hummed melody or basic piano line into a full orchestral arrangement.
• Use MIDI chord progressions to generate a vocal choir performance.
• Swap instruments in a track using text instructions while keeping the original melody intact.
• Access the Lyria RealTime API, which operates on a chunk-based autoregression system over a bidirectional WebSocket connection, generating audio in two-second chunks with live user control.

This is where Google’s technology starts to look genuinely impressive on a technical level. The model generates audio at 48kHz, maintains long-range musical coherence across the full duration of a track, and uses cross-modal embeddings to accept steering from text, images, or audio inputs simultaneously. That’s not a trivial engineering achievement, and it positions Lyria 3 well ahead of where the open-source alternatives currently sit.

@musictech_weekly on X (Feb 19, 2026): “Spent the morning with Lyria 3 in Gemini. The auto-lyrics are better than expected and the genre range is genuinely wide. The 30-second cap is frustrating but for Shorts content this thing is basically plug-and-play. Google figured out the use case.”

The Bottom Line

Google Lyria 3’s integration with Gemini and YouTube’s Dream Track gives it distribution advantages that standalone competitors like Suno and Udio simply don’t have. Thirty seconds is still a limitation, and the editing toolset is thin compared to Suno Studio. But for creators who live in short-form video, the tool is already fit for purpose. SynthID watermarking and multilingual support show that Google is building with scale and accountability in mind. This is a first chapter, not a finished product, but it’s a strong one.