Google adds AI music generation to Gemini app using DeepMind model

Google adds music generation to Gemini

Google launched a music-generation feature for its Gemini app on Wednesday. The tool uses the Lyria 3 model from Google DeepMind to turn text descriptions into audio tracks with lyrics and cover art. This beta release marks Google's most direct attempt to compete with standalone AI music apps like Suno and Udio.

Users can prompt the app to create specific musical scenarios, such as a comical R&B song about a lost sock. Gemini then produces a 30-second track and generates accompanying cover art via Nano Banana. The system handles the composition, vocal performance, and lyrical arrangement simultaneously.

The feature supports several languages at launch to reach a broad user base. Google confirmed the tool is available to users aged 18 and older. Supported languages include:

• English
• German
• Spanish
• French
• Hindi
• Japanese
• Korean
• Portuguese

Lyria 3 improves musical complexity

Google DeepMind developed Lyria 3 to produce more realistic and complex musical structures than previous iterations. The model allows users to exert granular control over the output by adjusting specific parameters. You can modify the style, vocal characteristics, and tempo of a track after the initial generation.

The tool also accepts multi-modal inputs beyond simple text strings. Users can upload a photo or a video file to the Gemini app to set the tone for a song. The AI analyzes the visual content and generates a soundtrack that matches the perceived mood of the media.

This integration reflects Google's broader strategy to weave generative media tools into its existing productivity suite. By placing music generation inside Gemini, Google bypasses the need for users to visit specialized third-party websites. The company aims to make AI music creation a standard part of the mobile assistant experience.

YouTube Dream Track goes global

Google is expanding its Dream Track feature for YouTube creators alongside the Gemini update. This tool also utilizes the Lyria 3 model to help creators build custom soundtracks for their content. Previously, Dream Track remained limited to a small group of creators within the United States.

The global rollout gives YouTube creators worldwide the ability to generate AI music for Shorts and long-form videos. This expansion suggests Google is confident in the model's performance and its legal safeguards. The company wants to normalize AI-assisted scoring for social media content.

Dream Track serves as a bridge between professional music production and casual content creation. It allows creators to avoid copyright strikes by generating original, licensed-by-default background music. This move could significantly shift how independent creators source audio for their videos.

SynthID watermarks all AI audio

Google is implementing strict identification measures for all music created with Lyria 3. Every song generated by the model includes a SynthID watermark. This digital identifier remains embedded in the audio even if the file undergoes compression or minor editing.

The company is also adding a detection tool directly into the Gemini interface. Users can upload an audio file and ask the app if it contains a SynthID watermark. This provides a way to verify whether a specific track originated from Google's AI tools.

These safety measures target the growing concern over deepfakes and unlabelled AI content. By providing a verification loop, Google attempts to position itself as a responsible actor in the generative AI space. The SynthID technology works by subtly modifying the audio frequency in a way that is inaudible to humans but readable by software.

Filters prevent direct artist mimicry

Google explicitly designed Lyria 3 to avoid the direct imitation of existing musical artists. The app includes filters that block prompts requesting a specific singer's voice or likeness. If a user includes a famous artist's name in a prompt, Gemini treats it as a broad stylistic suggestion rather than a directive to clone them.

The company stated that its goal is to encourage original expression rather than copyright infringement. The system checks outputs against a database of existing content to ensure the AI does not reproduce protected melodies. This approach contrasts with some competitors that have faced criticism for allowing "in the style of" prompts that closely mirror specific performers.

Despite these filters, the line between "inspiration" and "mimicry" remains thin in the world of generative audio. Google has not disclosed the full extent of the training data used for Lyria 3. However, the company has established licensing deals with major labels like Universal Music Group to explore legal monetization models.

Legal battles shadow AI music

The launch of Gemini's music features comes at a volatile time for the AI industry. Major record labels, represented by the RIAA, are currently suing AI music startups like Suno and Udio. These lawsuits allege that the companies trained their models on copyrighted recordings without permission or compensation.

Google is attempting to navigate this landscape by working directly with the music industry. By signing contracts with labels, Google and YouTube aim to create a system where AI-generated music generates revenue for rights holders. This strategy seeks to avoid the massive legal liabilities currently facing other AI firms.

The music industry remains divided on how to handle this technology. While some platforms like Spotify and YouTube are leaning into AI tools, others are focused on protection. Deezer, for example, recently launched tools specifically designed to identify and remove AI-generated music to prevent "fraudulent streams" from diluting the royalty pool for human artists.

The future of generative sound

Google's decision to bring Lyria 3 to Gemini suggests the company sees music generation as a core utility for AI assistants. The ability to generate a custom soundtrack for a video or a personalized song for a friend adds a layer of creative functionality to the app. It moves Gemini away from being a simple text-based chatbot toward a comprehensive media creation engine.

The success of this feature will likely depend on the quality of the 30-second clips. If the audio sounds too synthetic or "robotic," users may stick to existing libraries of royalty-free music. However, the convenience of generating a track in seconds via a mobile app is a significant advantage for casual users.

As the beta progresses, Google will likely expand the duration and complexity of the generated tracks. For now, the focus remains on short-form content and social media sharing. The global rollout ensures that Google will gather massive amounts of user data to further refine Lyria 3 for future versions.