FlowtandBriefAI · Software · ResearchMeet Flowtand

Google launches Lyria 3.5 in Flow Music with wider creative controls

Announcement

Google DeepMind

  • Models
  • #lyria
  • #music generation
  • #flow music

Browse Models topic page →

Google announced Lyria 3.5, a new music generation model rolled out in Google Flow Music. The company says it improves musicality, lyrics and vocal quality, and adds easier control over tempo and duration. The announcement is descriptive and offers no benchmarks, pricing or independent verification.

What Google announced

Google says it is launching Lyria 3.5 in Google Flow Music, describing advances across musicality, lyrics, vocals and creative control. Specifically, the post claims richer and more complex melodic structures, lyrics with better prompt adherence and structural awareness, vocals with more expression and improved pronunciation, and easier control of tempo and duration. The announcement is a product update rather than a technical report: it names features but gives no comparative listening tests, no quantitative measures and no independent results. What the evidence supports is the existence of the rollout and Google's own description of the intended improvements.

A scenario for a jingle producer

A producer at a small agency could trial Lyria 3.5 in Flow Music when drafting short commercial jingles, where tempo and length must fit a fixed slot. The claimed tempo and duration controls would be the relevant features to probe first, since they map directly to brief constraints, while the lyric and vocal claims would matter for anything with vocals. A team could try generating a handful of candidate sketches for one brief and comparing them with its usual drafting route. This is a hypothesis to test, not a claim that the model meets broadcast or client standards.

How to run a fair trial

A team could pick one real brief and one baseline method, then produce a small set of candidates with Lyria 3.5 and the same number with the existing process. Useful measures include how many drafts meet the brief's tempo and length without manual editing, how often lyrics follow the prompt, how many vocal takes are judged usable, and time spent per usable draft. Failure checks should include off-brief tempos, lyric drift and mispronunciation. Since the source provides no benchmarks, keep quality judgments with named human reviewers and record disagreements rather than relying on impressions alone.

Trade-offs and when to hold back

The main trade-off is speed of drafting against the editorial work of judging output quality, which the announcement says nothing about. Prerequisites include access to Flow Music and reviewers able to assess musicality and pronunciation. Unknowns include pricing, licensing terms for generated tracks, data handling and whether the features behave as described outside the vendor's framing. A creator should not rely on it for final commercial releases, sensitive brand vocals or work needing documented rights until those questions are answered. If the brief demands a specific artist style or strict compliance, keep the existing process as the reference.

Context and limitations

  • The evidence is a short announcement excerpt; the full article and any technical details were not reviewed.
  • No benchmarks, listening tests, pricing or independent evaluations are provided to substantiate the quality claims.
  • Tempo and duration controls are described only in general terms, without interface or API details.
  • Based on bounded source text, not independent testing. Practical suggestions are editorial proposals.