Google has unveiled a novel feature within its Gemini application: the ability to generate original 30-second songs using artificial intelligence. Powered by Lyria 3, Google DeepMind’s most advanced music generation model to date, this tool allows users to create musical tracks from simple text prompts, uploaded photos, or videos. The rollout began on February 18, 2026, and is currently available globally to Gemini users aged 18 and older, initially on desktop platforms with mobile access expected in the coming days.
The integration of Lyria 3 marks a significant step in the democratization of music creation, offering a low-barrier entry point for individuals to express themselves through sound. Unlike previous iterations of the technology, Lyria 3 automatically generates lyrics based on the user’s input, eliminating the need for manual lyric writing. Users also gain greater control over stylistic elements, including vocals and tempo, resulting in more complex and realistic musical compositions. Google emphasizes that the model is designed for original expression, and safeguards are in place to prevent the direct replication of existing artists’ work.
This isn’t Google’s first foray into AI-powered music generation. Lyria was initially launched in 2023 and subsequently utilized in experimental YouTube features like Dream Track, which enabled creators to generate short background music clips for YouTube Shorts. According to Music Business Worldwide, the new Lyria 3 model represents a substantial improvement over its predecessor, offering enhanced capabilities in lyric generation, stylistic control, and overall musical complexity.
How Lyria 3 Works: From Prompt to Song
Lyria 3’s functionality centers around its ability to translate user input into a complete musical piece. Users can initiate the process by providing a text prompt describing the desired genre, mood, or theme. For example, a user might request “a fun afrobeat track about childhood memories and plantains,” as demonstrated in a Google example. 9to5Google reports that the system can also interpret visual cues, allowing users to upload photos or videos as inspiration for the generated track. The result is a 30-second song complete with AI-generated lyrics and a custom cover art image created using Google’s Nano Banana tool.
To illustrate the capabilities of Lyria 3, one user tested the system by requesting a rap song in the style of 50 Cent about overcoming heartbreak and striving for a better life. As demonstrated on Ghacks, the resulting track, titled “Empire of Ashes,” showcased the model’s ability to generate both musical composition and lyrical content based on a specific stylistic prompt. You can listen to the generated track here:
Copyright Considerations and Responsible AI Development
Recognizing the potential for copyright concerns surrounding AI-generated music, Google has implemented several safeguards within Lyria 3. The company states that the model is specifically designed to foster original expression rather than directly mimicking existing artists. If a user includes a specific artist’s name in a prompt, the system will interpret this as a request for stylistic inspiration, generating a track that shares a similar mood or genre without directly copying the artist’s work. Google has incorporated filters to identify and prevent the generation of outputs that closely resemble existing songs.
Yet, Google acknowledges that these safeguards are not foolproof and encourages users to report any content that may potentially infringe on existing copyrights. The company has been careful in its training data selection, stating it has been “particularly mindful of copyright and partner agreements” and utilizing music for which it possesses the necessary rights. As detailed in the Google Workspace Updates blog, this approach aims to mitigate legal risks and promote responsible AI development within the music industry.
Accessibility and Future Developments
Currently, Lyria 3 supports music generation in eight languages: English, German, Spanish, French, Hindi, Japanese, Korean, and Portuguese, with plans to expand language support in the future. The feature is integrated directly into the Gemini app’s “Tools” menu, making it readily accessible to users who have access to the application. Administrators can control access to the Gemini app and its associated tools, including music generation, through the Generative AI settings in the Workspace Admin console.
While the initial output from Lyria 3 is limited to 30-second tracks, the potential for future development is significant. The technology could be expanded to generate longer-form compositions, offer more granular control over musical parameters, and potentially integrate with other creative tools. The update to YouTube’s Dream Track, utilizing Lyria 3, suggests a continued focus on providing AI-powered music creation tools for content creators. Whether YouTube creators will eventually be able to generate tracks exceeding the 30-second limit remains to be seen.
Key Takeaways
- AI-Powered Music Creation: Google’s Lyria 3 enables users to generate original 30-second songs from text prompts, images, or videos within the Gemini app.
- Automatic Lyric Generation: The model automatically creates lyrics, simplifying the music creation process for users.
- Copyright Safeguards: Google has implemented filters and guidelines to prevent the generation of music that infringes on existing copyrights.
- Multilingual Support: Lyria 3 currently supports eight languages, with plans for future expansion.
- Accessibility: The feature is available to Gemini users aged 18 and older, with administrative controls for Workspace users.
Google’s launch of Lyria 3 represents a compelling intersection of artificial intelligence and musical creativity. As the technology continues to evolve, it has the potential to empower a new generation of musicians and content creators, while also raising important questions about copyright, originality, and the future of music itself. The next update regarding Lyria 3 is expected to come from Google DeepMind, potentially detailing further language support and feature enhancements, in the coming months. Share your thoughts on this new technology in the comments below.