Lyria 3.5 API: music controls and song revision limits

Lyria 3.5 gives creators more control over generated music. But its API does not support a back-and-forth edit of an existing song. A new request can produce a different performance.

That gap matters when the brief is precise. “Make a new song with a slower tempo” is different from “keep this song and fix one line.”

What controls did Lyria 3.5 add?

Google's July Flow Music announcement described improvements to lyrics, vocals, and musical quality. It also added controls for tempo and duration. Those quality claims come from Google, not from a listening test for this article.

The September Gemini announcement expanded the model's reach. It described choices for vocal or instrumental music and short or longer tracks.

How the API differs from a song editor

The current API documentation lists Lyria 3.5 for songs lasting a few minutes. It also lists a separate clip model for 30-second output. The API supports 44.1 kHz stereo audio and adds a SynthID watermark.

The same documentation says requests are single-turn. It does not support conversational refinement. Repeating the same prompt also does not guarantee the same song.

These are API facts. They are not a promise that every Google app exposes identical controls or access.

Impact on users

Our analysis: More control can make a first draft closer to the brief. The limit appears during revision. A creator can like the vocal on one result and the arrangement on another, without a documented way to merge them.

That affects time as well as money. A release plan that assumes one small prompt edit will preserve the rest of a song can fail. Each new result can require another review of lyrics, timing, and structure.

A lyric video makes this problem visible. If the song changes after the words are timed, the video needs more work. A stable audio master matters more than a promising early sample.

How to plan a release around single-turn generation

Write a clear brief before the first request. Save the results that matter. Choose the final audio before you time lyrics or build a release video.

If you need exact local edits, keep that requirement separate from whole-song generation. Use an audio editor for changes that must preserve the rest of a performance.

What this report does not establish

This article did not compare sound quality, measure generation speed, or calculate a cost per finished song. It also does not treat a provider's permission to use output as a full answer about music rights.

The next useful test is a real revision task: keep the song, change one section, and count what else changes. A feature list cannot answer that question.