Google switched on its newest music generation model inside the Gemini app and API on September 4, 2026, giving 950 million monthly active users and any developer with an API key the ability to produce full-length songs from text prompts. Every track ships with an imperceptible SynthID watermark, satisfying a provider-side marking obligation under the EU AI Act but leaving the deployer-side labelling question unanswered for advertisers and agencies who publish the output commercially.
What the model produces
Lyria 3.5 generates songs lasting a couple of minutes with distinct verses, choruses and bridges, returning 44.1 kHz stereo audio as MP3 or WAV. The consumer interface inside Gemini adds genre selection, a vocal or instrumental switch, templates for use cases from background music to custom birthday tracks, and adjustable track length. Google named three commercial applications in its announcement: a backing track for a video, a brand jingle, and a personalised ringtone. Two of those sit inside marketing production, categories that have historically required a licensing conversation.
The developer surface exposes two models. Lyria 3 Clip returns exactly 30 seconds and is designed for loops, previews and prompt testing. Lyria 3.5 handles the full song. Both accept up to ten images alongside a text prompt, meaning a campaign key visual or product photograph can become the brief for an audio bed. Prompting follows a production brief structure: genre, instruments, beats per minute, key, mood, structural tags such as [Verse] and [Chorus], and duration. Timestamp control lets a generated cue land on a specific cut rather than near it.
The model works through musical structure before generating, using an internal prompt rewriter that is not exposed to the user. Intermediate reasoning blocks and thought signatures are withheld. Music generation is single-turn, results are non-deterministic even with identical prompts, and safety filters block requests for specific artist voices or copyrighted lyrics. Language handling follows the prompt, with lyrics generated in the language the prompt is written in and vocal style adapting accordingly.
The watermark and the disclosure gap
Every Lyria 3.5 track carries a SynthID audio watermark. DeepMind introduced the tool in 2024, and Google has since threaded it through multiple products: consumer-facing video verification arrived in the Gemini app in December 2025, and Gemini 3.5 Live Translate shipped with the watermark woven into generated speech in June 2026. The marker is designed to survive editing and reformatting, and Google says it does not affect the listening experience.
The regulatory context changed sharply before this release. Article 50 of the EU AI Act became legally applicable on August 2, 2026, with penalties reaching 15 million euros or 3 percent of worldwide annual turnover. The European Commission published its implementing guidelines and a finalised Code of Practice on Transparency of AI-Generated Content on July 20, 2026. Providers face machine-readable marking duties. Deployers - a category that includes advertisers, agencies and publishers - face labelling duties where the content qualifies. An imperceptible watermark satisfies the provider obligation. It does not, by itself, satisfy a deployer obligation that requires a visible disclosure.
Google has been building the advertiser-facing half of that stack separately. On July 9, 2026, the company introduced a "How this ad was made" disclosure panel across Google Ads, Display and Video 360, Campaign Manager 360, Merchant Center and Ads Editor, with automatic disclosure for assets produced in Google's own tools. A track generated in the consumer Gemini app and uploaded by hand does not travel through that pipeline. Nothing in the Gemini app interface answers the labelling question at the point of generation.
The licensing environment
Generated music is not arriving into a settled legal picture. On July 31, 2026, the 42nd Civil Chamber of the Munich I Regional Court ruled against Suno in a case brought by the German collecting society GEMA, setting a penalty of 250,000 euros per breach. The 143-page judgment applied United States copyright law and concluded that fair use did not shield the training. Universal Music Group settled its litigation with Udio in October 2025 and agreed licences for a walled-garden platform, a structure an artist coalition challenged on transparency grounds the following month.
Rights holders have begun applying sync-licence theory to ordinary brand social output. Sony Music filed a 58.8 million dollar claim against Kroger in August 2026 covering 392 songs used across the retailer's social accounts, citing 14 prior Sony licences since 2017. A generated track carries no rights-holder counterparty. What it carries instead is uncertainty about the training data behind it, which is the question the Munich court answered against Suno. Platforms have started building disclosure machinery on the distribution side: Apple Music introduced Transparency Tags on March 4, 2026, placing disclosure duty on labels and distributors with no published cross-verification mechanism.
The distribution scale
Lyria has been threaded through Google properties for close to three years. YouTube ran Dream Track as a limited experiment from late 2023, generating Shorts soundtracks in the voices of nine participating recording artists. In September 2025, the platform integrated Lyria 2 through a Speech to Song feature. By mid-2026 the model had reached third-party advertising software: Typeface added Lyria-powered soundtrack generation to its Video Agent product in July 2026, with the system reading a video's duration and tone before producing a matching track.
The Gemini app placement changes the volume question rather than the capability question. Alphabet reported 750 million monthly active users for the Gemini app in its fourth quarter 2025 results. The figure crossed 900 million by May 2026 and reached 950 million in the second quarter of 2026, with daily actives tripling year over year. Placing a music generator inside that installed base removes the scarcity of finished audio for the low end of the market. The bespoke commission does not disappear. The 200-dollar library cue for a fifteen-second social edit has a harder case to make.
The rollout was staged across multiple dates. Google published a separate post on July 29, 2026 announcing Lyria 3.5 in Google Flow Music. The Gemini API changelog lists the model in public preview from September 3, 2026. The September 4 consumer post presents the model as newly available without referencing the earlier Flow Music release. Coverage that treats either date as the launch describes half of the story.
Why this matters for creatives
Three practical consequences follow. The first is supply. A production music library, a stock music subscription and a bespoke composer commission all price against scarcity of finished audio. A model that returns a structured two-minute track with vocals from a text prompt, at no cost inside a consumer application reaching close to a billion monthly users, removes that scarcity for quick-turn social and digital work. The second is compliance. The watermark answers a provider obligation. It does not answer the deployer obligation that falls on the agency or the brand, and it is invisible by design, which means no reviewer catches it by listening. Whether a generated track requires a visible label depends on the application, the legal role of the party publishing it and the content type. The third is verification. A track produced in Lyria 3.5, exported, re-encoded and delivered to a supply-side platform is identifiable only if a downstream system looks for SynthID. Buyers have not adopted that signal into brand safety classification, and the infrastructure for doing so exists on some platforms and not others. For creative professionals working across AI for Creatives and Generative Video, the tool is now in their pocket. The disclosure duty is now on their desk.
Your membership also unlocks:
