Gemini just
learned to compose.
Until now, "making music with AI" meant reaching for a tool outside Gemini. The Lyria 3.5 music model has been built directly into the Gemini app, so a text prompt is now enough to get a background track or jingle. Here is what actually changed, and who it is for.
Text and images are done.
Video and audio are next.
The democratization of generative AI has always followed the same order: text first, then images. Both are now commodities anyone can use, and the standalone apps built around them lost most of their edge. What was left was audio and video — and this week it was Google's music model, Lyria, that moved.
Lyria used to live mostly inside dedicated tools: a music feature in the "Google Flow" video editor, and a capsule feature inside the Gemini app. Google's own blog post describing the custom-soundtrack feature in the Gemini app is a good example of that earlier framing — Lyria 3 as an in-app generator, not a platform-wide capability.
That changed on September 4, 2026, when the newer Lyria 3.5 rolled out simultaneously to the Gemini app, Gemini API, Google AI Studio, and Google Vids, becoming available globally to every Gemini user on web and mobile. Promoting a dedicated tool's feature into a general-purpose assistant's standard capability is exactly the pattern image generation went through a few years ago, moving from "a separate service" to "a feature inside the chat." The same commoditization wave that hit text and images now appears to be reaching music.
| Before | After Lyria 3.5 |
|---|---|
| Composing meant a separate tool (e.g. Flow Music) | Compose directly from the Gemini app, by text |
| Lyria 3 stayed a Gemini app-only feature | Lyria 3.5 spans the app, API, AI Studio, and Vids |
| Mainly framed as soundtrack generation | Dedicated templates for BGM, jingles, ringtones |
How Lyria got here
From a Flow Music experiment to the Gemini app: a "niche feature" became a "standard feature" in a bit over half a year.
According to Unite.AI's reporting, Google calls Lyria 3.5 its most expressive music model to date, with richer vocal expression, fuller musical arrangements, and higher audio fidelity than earlier Lyria versions. Tracks are not simple loops — they can carry a structure of verse, chorus, and bridge — and the stereo, up-to-3-minute ceiling keeps the length gain Lyria 3 Pro delivered back in April while raising quality on top of it.
On the Gemini app side, the new feature set includes three purpose-built templates — background music, brand jingles, and personalized ringtones — with genre selectable from presets or described in free text. Users can also toggle between vocal and instrumental (no-vocal) versions, which speaks directly to a practical need: producing a "song" version and a "background-only" version from the same brief.
Who benefits, and how
The biggest gains go to people who need sound but are not music specialists.
Designers: draft the mood
Prototype the feel of a video or presentation on the spot, without commissioning a composer or hunting through stock-music licenses. At the storyboard or internal-pitch stage, this level of quality is already enough to move the conversation forward.
Marketers: cheap jingle iteration
Generate several brand-jingle or ringtone variations per campaign at low cost, compare genres and moods, and only then brief a composer if one is needed. It lets you produce your own "rough cuts" before the creative brief exists.
The draft can now be in-house.
The finish is still in human hands.
What to try first
Start with private drafts
Try it first on mood boards, internal-only background music, or a rough jingle for a pitch — anything that stays out of public release — to get a feel for what genres it handles well and where it struggles.
Check terms and disclosure rules before going public
Before using output in a paid campaign or public release, confirm Google's usage terms and any requirement to disclose AI generation or attach provenance metadata. Being able to make a track is not the same as being cleared to ship it commercially.
Keep a human path for high-stakes work
For a brand's signature theme, or any release that needs cleared rights end to end, a professional composer or music production house remains the realistic choice.
Why not to over-trust it
Google itself frames this as not a replacement for professional composition. It widens how much background music and jingle work can be brought in-house, but it is not yet a substitute for release-grade music production.
Broader unresolved questions around AI-generated music also still apply in 2026. One is licensing ambiguity — how independent a generated melody really is from copyrighted material in the training data is hard for an outside party to verify. The other is genre-level sameness: preset labels like "Lo-fi" or "Cinematic" alone tend to converge on generic-sounding output without deliberate prompt craft. That may be fine for background use, but it can fall short for a piece meant to represent a brand.
In practice, treat this as a capable drafting and prototyping tool — the final responsibility for rights clearance and finished quality still sits with the person using it.