Gemini Lyria 3 music: Google launches 30-second track generator

Google's Gemini app now includes Lyria 3, DeepMind's newest music generation model, enabling users to produce 30-second musical tracks from text or image prompts. The feature is available in the Gemini app and is positioned as an easy entry point for anyone to create short audio clips without composing or recording manually.

Gemini users can type descriptive prompts or supply images to steer Lyria 3's output. The model returns compact, 30-second stereo audio files intended for quick sharing and iteration. DeepMind describes Lyria 3 as its most advanced music model to date and integrates it into Gemini's existing chat and creative toolset.

How Lyria 3 generates music in Gemini

Lyria 3 accepts natural language prompts and image inputs and produces a short audio clip that matches the requested mood, genre, instrumentation, and tempo. Users can specify elements such as "cinematic piano with soft strings" or upload a photograph to influence the track's atmosphere. The Gemini interface presents simple sliders and options to adjust style and to regenerate variants.

Under the hood, Lyria 3 synthesizes melody, harmony, rhythm, and timbre into a single audio output. Generation typically results in a 30-second stereo file ready for download or sharing. Where device capability allows, some processing may occur on-device; otherwise creation is handled in the cloud to accommodate computational demands and larger model sizes.

The Gemini app surfaces generated audio alongside a textual explanation of the prompt and a set of alternative variations. Users can iterate by modifying prompts or requesting different instrumentations. Export options include direct sharing from the app and saving local copies of the WAV or MP3 files.

Controls, limitations, and policy guardrails

DeepMind and Google emphasize user controls and safety filters around copyrighted material and explicit content. The model includes mechanisms to avoid generating audio that directly imitates identifiable copyrighted songs or known artists, and the app applies content policies to prevent disallowed outputs. Users are prompted to confirm rights for any uploaded images used as prompts.

Audio length is constrained to short clips to keep generation fast and to limit potential misuse. The company provides guidance on responsible use and labels generated tracks as AI-created. Commercial use and licensing terms for outputs are covered in the app's terms of service; creators who need broader rights are directed to the platform's developer offerings and licensing documentation.

Early examples showcased by DeepMind include ambient pads, lo-fi beats, orchestral stabs, and short vocal textures. The feature targets hobbyists, social creators, and teams looking to prototype musical ideas quickly rather than replace full-scale composition workflows.

Why it matters

Embedding Lyria 3 in the Gemini app lowers the barrier to creating original music by letting users convert words and images into immediate audio examples. Short, iterative clips speed experimentation for creators and streamline the early stages of music production and sound design. Wider rollout of models like Lyria 3 will sharpen debates over copyright, attribution, and how AI-generated audio should be licensed and used in creative work.