Google announced Lyria 3.5 in the Gemini app and Gemini API on September 4, 2026. The company describes more expressive vocals and richer arrangements, with controls for genre, vocal or instrumental output, templates and shorter or longer tracks. It also names access through Flow Music, AI Studio and Google Vids.
This is a dated September development, examined here on October 3 rather than recast as breaking news. The quality descriptions are Google’s claims. AI Speed has not performed a listening comparison, and access through one product should not be assumed to have identical controls or terms in another.
Write a brief that can guide revision
A useful creative brief specifies the role of the music: background under narration, a transition or a foreground song. Describe mood, duration and whether a voice is needed. Those choices help you assess the result against the project instead of accepting the first track merely because it sounds complete.
For a Canadian creator making bilingual material, listen closely to any generated pronunciation and phrasing. Have someone fluent in the intended language check the lyrics and meaning. An attractive melody can distract from an awkward line or an unintended message, especially when the words are heard quickly in a short video.
Related sources: Google Lyria 3.5 Gemini announcement ↗ · Google DeepMind Lyria model resource ↗
Listen in the actual context
Place the candidate track under the intended narration or visuals before judging it. Check whether the arrangement obscures speech, whether the ending fits the cut and whether repeated sections become distracting. Test at ordinary listening levels and on the kind of device the audience will use, rather than only through studio headphones.
Keep a revision record: input, chosen product, date, output version and edits. That helps collaborators understand why a track was selected and makes later changes easier. If several people work on a project, agree on which version is approved so a rough experiment does not accidentally become the final soundtrack.
Distribution requires current terms
Review the applicable service terms and the destination platform’s requirements before publishing commercially. Keep permission records for any uploaded reference material and avoid implying that a generated performance was made by a real artist who did not participate. Availability of a generation tool is not a complete answer to a project’s licensing questions.
Our analysis is that the release gives creators more ways to explore a musical idea, while leaving judgment about fit and presentation with the creator. The strongest use is a deliberate process: define the brief, generate options, listen critically and document the final choice. More expressive output does not eliminate the need for that review.
Sources and editorial method
We use the linked original statements and reports to distinguish documented facts from our analysis. Company announcements describe their authors’ plans and claims; they are not independent verification of future outcomes. This article is dated and will be revised if material evidence changes. Read our editorial standards and corrections process.
- Google Lyria 3.5 Gemini announcement ↗ — Primary September 4, 2026 announcement; supported creative controls and stated product availability.
- Google DeepMind Lyria model resource ↗ — Official model resource; consult current documentation and terms for the chosen product and intended distribution.