Google DeepMind has rolled out Lyria 3, its latest generative music model, inside the Gemini app, marking the company’s next step in AI-powered creative tools. The feature allows users to generate 30-second original music tracks from text prompts, photos, or videos within seconds.
The update expands Gemini’s creative capabilities beyond images and video, adding custom music generation designed for casual expression and social sharing.
Text-to-Music and Image-to-Track Functionality
Users can create a track by describing a genre, mood, memory, or concept. For example, prompts such as a nostalgic Afrobeat tribute or a humorous R&B slow jam can be converted into short, fully produced songs complete with lyrics.
The tool also supports uploads. Photos and videos can be transformed into music, with Gemini generating lyrics and instrumentals that reflect the mood or subject of the content.
Each track is accompanied by automatically generated cover art created using Nano Banana and can be downloaded or shared via link directly from the app.
Improvements in Audio Quality and Creative Control
Lyria 3 builds on earlier versions of Google’s music models with several upgrades:
-
Automatic lyric generation: Users no longer need to supply lyrics.
-
Greater creative control: Prompts can specify elements such as style, tempo, and vocals.
-
More realistic and complex compositions: The model produces more layered and musically detailed tracks.
The company says the goal is not to replace professional music production, but to offer a playful way for users to create personalized soundtracks.
Integration With YouTube Shorts
YouTube creators can also experiment with Lyria 3 through Dream Track, a feature designed to enhance soundtracks for Shorts. Initially launched in the United States, the feature is expanding to creators in additional countries.
The update enables users to generate lyrical verses or backing tracks tailored to their short-form videos, offering more customization for content creators.
AI Content Verification and SynthID Watermarking
All tracks generated in Gemini are embedded with SynthID, Google’s imperceptible watermarking technology for identifying AI-generated content.
The Gemini app also introduces expanded verification capabilities for audio. Users can upload a file and ask whether it was generated using Google AI. Gemini will check for SynthID markers and apply its own analysis before returning a response.
Copyright Safeguards and Responsible Development
Google states that Lyria 3 was developed in collaboration with the music community, following earlier experiments such as Music AI Sandbox. The company says it has considered copyright and partner agreements during the training process.
Music generation is intended for original expression rather than direct imitation of existing artists. If a prompt references a specific artist, Gemini interprets it as broad stylistic inspiration rather than replication. The system also includes filters to compare outputs against existing works.
Users can report potential rights violations, and all usage is subject to Google’s Terms of Service and generative AI policies.
Availability
Lyria 3 is available in the Gemini app to users aged 18 and older in multiple languages, including English, German, Spanish, French, Hindi, Japanese, Korean and Portuguese. Expanded language support is planned.
