Can Gemini Make Music?

We've now written this article three times — for Grok, for ChatGPT, and now for Gemini — and Gemini is the first one where the answer isn't "no." That alone makes it worth being precise about, because "Gemini can make music" and "Gemini can make your song" are two different claims, and only one of them is true. We build an AI music app, so consider our bias declared — and our testing obsessive.

The 60-second answer

• Yes, Gemini genuinely generates music. Since early 2026 the Gemini app has a Create Music tool powered by Lyria 3, Google DeepMind's music model. Type a prompt — or hand it an image — and it returns a real, listenable track.
• The free tier makes 30-second clips. Short mood pieces, not songs. On Gemini's paid tiers, Lyria 3 Pro stretches that to tracks up to 3 minutes, with vocals and timed lyrics.
• Every track is watermarked and the rights are murky. All Lyria 3 audio carries Google's inaudible SynthID watermark, and commercial permission from Google is not the same thing as owning a copyright.
• It's a soundtrack tool, not a song studio. For a track you shape and own creatively — your structure, your voice, section-by-section edits, a release path — a dedicated AI music app is still the right tool.

Gemini can hand you a track. What it can't hand you is your track — the one you shaped, edited, and put your own voice on.

What Lyria 3 actually delivers

Credit where due: Lyria 3 is the real thing, not a demo. It generates high-quality 44.1 kHz stereo audio with genuine structure — intros that build, sections that change, vocals with lyrics timed to the music. You can prompt it with text or with an image (a photo of a rainy street becomes a rainy-street soundtrack), and you can remix a track you've already made. In the Gemini app it's as simple as picking "Create music" from the tools menu; it rolled out globally in eight languages, for users 18 and up.

The tiering is what most coverage glosses over. The standard experience — what you get free — produces 30-second tracks. That's a hard shape: enough for a reel, a mood sketch, a background bed. The 3-minute capability people quote comes from Lyria 3 Pro, which launched in March 2026 and sits behind Gemini's paid "Pro/Thinking" tiers. So whether "Gemini makes full songs" is true depends entirely on which Gemini you're paying for — and even Pro's three minutes comes with the control limits below.

Quality-wise, it's impressive and improving, though reviewers have been pointed about the gap that remains — Engadget memorably called the output an approximation of what real music sounds like. That's roughly where we'd put it too: excellent for functional music, still short of a track you'd defend as yours.

The fine print nobody reads

The watermark. Every Lyria 3 track is embedded with SynthID, Google's inaudible identifier for AI-generated content. It's woven into the audio at creation and survives common edits — MP3 compression, speed changes, added noise. You can't hear it, but any platform that checks for it will always know the track is AI-made and where it came from. Depending on your use case that's either reassuring or a dealbreaker; either way, you should know it's there.

The rights. Google offers indemnification for covered generative AI services, which sounds like ownership but isn't. Permission to use a track commercially under a platform's terms is a license that lives and dies with those terms — and separately, US copyright law doesn't protect purely AI-generated output at all, a problem every AI music tool shares (we cover the ownership question in depth in our AI music monetization guide). If you're planning to put Lyria 3 audio in anything commercial, read Google's current terms first, not a blog post — including this one.

Where it stops

Here's the part that matters if what you want is a song rather than a soundtrack. Generating a track is the first third of making music you'd actually release. After that comes shaping it — and that's where Gemini's music tool runs out of road:

• No section-level control. If the chorus doesn't land, you regenerate and hope. There's no "keep the verse, redo the chorus," no extending a section, no swapping one instrument.
• No voice of yours. Lyria 3 sings in its own generated voices. There's no cloning your voice so the song actually sounds like you — a feature that's core to dedicated music apps (done with consent gates, in ours).
• A soundtrack workflow, not a release workflow. Gemini hands you a clip inside a chat app. It has no concept of building a track for TikTok's first two seconds, pairing it with a music video, or exporting stems for a distributor.
• The 30-second default. Unless you're paying for Pro, every idea comes back as a half-minute sketch.

None of this is a flaw, exactly — it's scope. Google built a delightful general-purpose feature inside an assistant. It did not build a music studio, and it doesn't claim to have.

So: Gemini or a dedicated music app?

Honest sorting, the way we'd tell a friend:

Use Gemini when the music is furniture. A quick bed under a video, a mood sketch while you're brainstorming, a "what would this photo sound like" toy — Lyria 3 is genuinely great at this, it's right there in an app you already have, and 30 seconds is exactly the right length.

Use a dedicated app when the music is the point. A song for TikTok that needs a hook in the first two seconds (we wrote the guide), a birthday song in your own cloned voice, a track you want to iterate on until the chorus is right and then actually put somewhere — that's what purpose-built apps like Sonx, Suno, and Udio are for. Our step-by-step guide walks the whole workflow from one-line idea to exported track.

And the two combine better than they compete: sketch the vibe in Gemini if that's where you are, then rebuild it properly — full length, your structure, your voice — in a music app.

TL;DR

Gemini really does make music now — Lyria 3 gives it 30-second tracks free and up to 3 minutes on Pro, with vocals, image prompting, and remixing. Every track is SynthID-watermarked, commercial rights are a license rather than ownership, and there's no section editing, no voice cloning, and no release workflow. For quick soundtracks, use it happily. For a full song that's actually yours to shape, a dedicated AI music app is still the tool for the job.