Google is making music creation more approachable in Gemini with Lyria 3.5, new templates and clearer controls. Here is what the update offers creators, and where expectations need to stay realistic.
Google announced on September 4, 2026 that Lyria 3.5 is available in the Gemini app and Gemini API. The company says its updated music generation model improves vocal expression and musical arrangements, alongside new ways to choose a style and get started from a template. Google’s launch announcement
For someone making a video, preparing a presentation or exploring a song idea, the practical question is whether these changes make it easier to get a usable result. A convincing piece of music still needs to match the project’s mood, pacing and purpose.
What Lyria 3.5 brings to Gemini
Lyria is Google’s family of AI music generation models. In Gemini, users can describe a musical idea or provide a photo as inspiration, then generate a track with vocals or an instrumental arrangement. Google’s product page advertises tracks of up to three minutes, accompanied by cover art generated with Nano Banana. Gemini’s music generation overview
The September update combines changes to the model with changes to the interface. Google highlights more expressive singing and richer arrangements, while the app adds templates and easier selection of genre, vocals and track length. Google’s launch announcement
Those controls could make a meaningful difference for beginners. Choosing an instrumental track, for example, gives someone creating a narrated video a clear starting point. Templates could also help users discover the kind of musical direction they want before writing a detailed prompt.
Track length, access and downloads
Google’s current help documentation describes the following options:
| Item | What Google documents |
|---|---|
| Short track | Approximately one minute |
| Full track | Approximately two to three minutes; select the “Pro” model |
| Musical direction | Genre selection and a choice of vocals or instrumental music |
| Downloads | MP3 audio or an MP4 video with cover art |
| Access requirements | Age 18+, a personal or supported work/school Google Account, and Keep Activity enabled |
| Usage limits | Generation quotas apply; users can wait for a reset or upgrade their subscription |
Source: Google’s music generation help
Google describes the release as available globally on the web and mobile. Its product page clarifies that music generation is offered in countries where the Gemini app is available, for users aged 18 and over. Google’s announcement, Gemini’s availability details
The distinction between model capabilities and account access matters: a maximum duration does not tell you how many attempts your account allows. That becomes relevant when a project needs several variations before one fits.
How to turn an idea into a useful music brief
DeepMind describes Lyria as a system that responds to musical direction, including vocal style, tempo and acoustic preferences. It also supports users supplying their own lyrics or asking for lyrics around a theme. Google DeepMind’s Lyria overview
For a first experiment, think about the role the music needs to play. A track underneath a spoken explanation has different requirements from a song intended to hold the listener’s full attention.
A prompt to try for a technology explainer could be:
Create an instrumental electronic track for a one-minute video explaining an AI feature. Use soft synths, a steady rhythm and a curious, optimistic mood. Keep the melody understated so spoken narration remains clear, and finish with a gentle resolution.
This is an illustrative prompt, not a tested result. Its purpose is to specify the audience experience: space for speech, a consistent mood and an ending that suits the video.
After generation, listen with the narration in place. Check whether the music distracts from key sentences and whether the ending feels natural. Adjusting volume or trimming a section in an editor may still be part of the process.
Creators assembling the visual side can also consult Demystia’s comparison of AI video generation tools.
What remains to be evaluated
Google’s description of improved sound is a claim about its own model. The launch announcement does not provide a quantified comparison or an independent listening study establishing how large the improvement is. Google’s launch announcement
A useful evaluation would compare multiple outputs from the same brief, checking vocal clarity, musical structure, adherence to instructions and consistency across attempts. One impressive example would tell a creator less than repeated success on the kind of work they actually produce.
DeepMind itself acknowledges that capabilities still need improvement and advises users to check whether generated tracks match their intentions. It also says Lyria audio carries SynthID, an imperceptible watermark designed to help identify music created or edited using Google AI. Google DeepMind’s limitations and safety information
These are separate considerations: identifying a track’s AI origin does not evaluate its musical quality or suitability for a particular project.
Why integration into Gemini matters
Our reading of this release is that accessibility may be as consequential as the sound improvements. Someone already working in Gemini can experiment with music as part of an existing project. Templates and visible controls could reduce the effort required to make a first attempt.
For dedicated music tools, this suggests a competitive question: how much more control, consistency and editing flexibility can they offer beyond a convenient first draft? The announcement alone does not establish a quality ranking between services.
For creators, the most useful test is concrete: can Lyria produce a track that fits a real project within a reasonable number of attempts? The update provides more ways to express that brief. Its everyday value will depend on how reliably the output follows it.