Coreqm News

Lyria 3.5 Arrives in Gemini: Music Generation Gets More Creative Controls

Google's September 4 update adds new ways to shape tracks. Creators can evaluate structure and editability as carefully as first-listen quality.

By Coreqm ·

Updated

Research checked September 5, 2026. Source-based reporting and Coreqm editorial analysis.

The release

Google announced Lyria 3.5 on September 4, describing improved vocals and richer arrangements. The company says the model is available in the Gemini app and Gemini API. Its app update includes genre selection or description, vocal and instrumental styles, templates and a choice of shorter or longer tracks.

These are announced product features. This article does not present an independent listening benchmark or claim that every generated track is ready for commercial release.

What creators should listen for

Coreqm analysis: A striking opening can make a generated track feel successful before its structure has been tested. The more useful listening exercise follows the entire piece: how it introduces an idea, develops it and reaches an ending.

For a video or game prototype, consistency may matter more than novelty. A track needs to fit the scene's energy and leave room for dialogue or effects. A dense arrangement can sound impressive in isolation while making the actual project harder to mix.

A practical comparison method

Use a short creative brief and keep it consistent across several generations. Describe mood, intended use and the role of vocals. Then compare the outputs against the brief rather than choosing only the most surprising result.

Listen for abrupt transitions, unstable vocal delivery and an ending that feels disconnected from the preceding section. Record which prompt changes actually improve the result. This turns repeated generation into a learning process instead of a search for a lucky sample.

From prototype to published work

Before using a track publicly, creators should review the applicable service terms and any distribution requirements for their project. The announcement of a model does not settle every licensing or ownership question.

The update gives users more ways to communicate musical intent. Its value will be clearest when those controls help creators reach a usable arrangement with fewer revisions. That is a different and more practical measure than whether the first few seconds sound convincing.

Source

Official announcement