Google Adds Lyria 3 Music Generation to the Gemini App

Announcement Overview

Google said on February 18, 2026 that it is bringing Lyria 3, its latest generative music model from Google DeepMind, into the Gemini app in beta. According to the announcement, users can type a prompt or upload an image to generate a custom 30-second track in seconds. Google positions the feature as a lightweight creative tool for everyday expression rather than a full production studio workflow. Source: Google announcement.

What Lyria 3 Adds

Google highlights three functional upgrades in Lyria 3. First, lyrics can now be generated directly from prompts, so users do not need to provide lyrics manually. Second, prompt control has improved for style, vocals, and tempo. Third, the system produces more realistic and musically complex tracks than earlier iterations. The app also generates matching cover art for each output using Nano Banana, making sharing workflows faster.

The post describes two core creation paths: text-to-track and media-to-track. In text-to-track mode, users specify genre, mood, or narrative intent. In media-to-track mode, uploaded photos or videos become the context for the generated soundtrack and lyrics. Google explicitly says the goal is fast, personal expression and social sharing, not replacing professional music production.

Distribution and Trust Controls

Google also says creators can use Lyria 3 through YouTube Dream Track for Shorts, starting in the U.S. and expanding to additional countries. This ties Gemini-generated music workflows to broader creator distribution channels.

On safety and attribution, Google states that all tracks generated in the Gemini app are embedded with SynthID, its imperceptible watermarking technology for AI-generated content. The company is also extending Gemini verification features so users can upload audio and ask whether it was generated with Google AI. The response combines SynthID checks with model reasoning. Google further notes that prompts naming specific artists are treated as broad stylistic inspiration and that policy/reporting paths remain in place for potential rights violations.

Strategically, this release matters because it combines model capability, creator tooling, and provenance controls in one product launch. It signals that consumer AI music generation is moving from demos into integrated app ecosystems with built-in trust layers.