Google rolls out Lyria 3.5, a new music model with better vocals and lyrics

Google has released Lyria 3.5, the latest version of its music generation model, on the Flow Music platform. The update mainly improves vocal sound and lyric quality, the elements that until now most gave away the artificial origin of AI-generated tracks.

What the new model improves

According to Google, Lyria 3.5 produces richer and more complex melodic structures that sound more natural than in the previous version. The biggest change, however, is in the vocals: the model is meant to generate more realistic and emotionally nuanced voices with better pronunciation, previously one of the weakest points of generative music.

Lyric quality has also improved. Lyria 3.5 sticks more closely to the user's prompt and understands song structure, meaning it can distinguish between verse, chorus and bridge instead of generating one unbroken string of phrases.

Creative control in Flow Music

Flow Music, the platform running the new model, lets users edit individual sections of a track without reworking the whole thing: they can rewrite the lyrics, translate them into another language, change the instrumentation, or restyle the beat while keeping the rest of the composition intact.

Creators also get direct control over tempo, vocals, drums and bass, and track length can be set anywhere from 30 seconds to three minutes. That brings Lyria 3.5 closer to a digital audio workstation than a one-shot generator built from a simple text prompt.

The shadow of copyright lawsuits

The new model's launch coincides with mounting legal pressure on the entire AI music industry. In March 2026, independent musicians, including New York songwriter Sam Kogon and Los Angeles composer Magnus Fiennes, sued Google in an Illinois district court, alleging that the previous version of the model, Lyria 3, was trained on 44 million audio clips from YouTube without creators' consent.

Google's defense is that YouTube's terms of service already granted the company a broad, royalty-free license to create derivative works from content uploaded by users. The company filed a motion to dismiss the lawsuit in June 2026, and the case is still ongoing.

What this means for the industry

Google's rivals face similar disputes. Sony Music filed a second lawsuit against Udio this month, accusing the company of copying 30,000 recordings to train its models and seeking $4.5 billion in damages. Universal Music Group and Warner Music Group previously dropped their lawsuits against Udio and Suno after signing licensing deals, but Sony remains the only major label still suing both companies.

For Google, which has its own YouTube catalog and millions of hours of recorded content, the key question is whether the platform's terms of service are really enough of a legal basis for training commercial models. The outcome of the Illinois case could influence how other tech companies with their own video or streaming services approach building music models.

For Polish musicians and music publishers, the update means yet another, increasingly refined tool for quickly prototyping tracks, but also a reminder that the legal status of the training data used by major music models is still being disputed in American courts, not a settled standard.