Google Brings Lyria 3 to the Gemini App in Beta for 30-Second AI Music Generation

What Google Announced

In its 2026-02-18 post, Google said Lyria 3 is rolling out in beta inside the Gemini app. The release expands Gemini from text and image assistance into short-form music generation, where users can describe an idea or upload visual input and receive a generated track. The official description repeatedly frames output length as 30-second tracks, optimized for fast creative iteration rather than full studio production.

Google also describes a paired output flow: the app creates music and generates matching cover art in the same interaction. The article states that Nano Banana is used for this cover generation path. This packaging matters because it turns a single prompt into a shareable mini-asset, not just raw audio.

Safety and Verification Layer

A central part of the announcement is provenance controls. Google says all tracks generated in the Gemini app are embedded with SynthID, its watermarking system for identifying Google AI-generated content. The company also says Gemini’s verification capabilities are expanding to audio, alongside image and video checks, so users can upload files and ask whether they were generated with Google AI signals.

This is important operationally: the launch is not only about generation quality, but also about downstream detectability and policy enforcement. Google additionally reiterates terms and prohibited-use constraints around intellectual property and privacy.

Availability and Rollout Details

• Launch date in post metadata: 2026-02-18.
• Access scope: Gemini app users aged 18+.
• Languages listed: English, German, Spanish, French, Hindi, Japanese, Korean, Portuguese.
• Rollout sequence: desktop first, then mobile app over the next several days.
• Plan tiers: Google AI Plus, Pro, and Ultra receive higher limits.

The broader signal is that multimodal consumer AI products are moving toward end-to-end creative workflows with built-in provenance controls. The next practical questions are adoption scale, creator satisfaction at 30-second length constraints, and how watermark verification performs in real redistribution environments.

Source: Google Blog