Gemini app: Lyria 3 music generation (30s, SynthID)
Key Points
- Lyria 3 generates 30-second tracks
- SynthID watermarking on all outputs
- Text- or image-to-music with auto-generated lyrics
Summary
The Gemini app now includes Lyria 3 (Google DeepMind) to generate 30-second music tracks from text or images. Outputs include auto-generated lyrics, customizable style controls, and a Nano Banana cover image. All tracks are embedded with SynthID (an imperceptible watermark) and can be verified inside the Gemini app. This feature is rolling out in beta for users 18+ with platform and subscription-dependent limits.
Key Points
- Model and output: Lyria 3 produces 30-second tracks (lyrics or instrumental) from text prompts or uploaded photos/videos.
- Controls: prompt-driven control over style, vocals, tempo; Lyria 3 supports more realistic and musically complex outputs than prior versions.
- Cover art: custom 30-second track cover art is auto-generated by Nano Banana for easy sharing.
- Watermarking & verification: every generated track contains SynthID; Gemini can check uploaded audio for SynthID and report whether it was produced by Google AI.
- Availability & limits: beta in the Gemini app for users 18+; supported languages include English, German, Spanish, French, Hindi, Japanese, Korean, and Portuguese; desktop rollout now, mobile follows, and subscribers (Google AI Plus/Pro/Ultra) get higher quotas.
- Responsible use: designed for original expression (artist names used only as inspiration); content filters check against existing works; reporting mechanisms and Gen AI policy/ToS apply.
- Integrations: Lyria 3 is also available for YouTube Dream Track (Shorts soundtracks) to improve creator audio quality.
Practical notes for engineering teams
- Expect short, shareable, multimodal audio assets (30s) with embedded metadata (SynthID) rather than long-form production-ready tracks.
- Verification is performed inside the Gemini app via file upload and SynthID detection; consider this when designing ingestion or moderation workflows that need AI-origin detection.
- Use cases are geared toward quick creative expression and social sharing; the feature is in beta and subject to rollout, limits, and policy constraints.