Google's Lyria 3 AI music model launches in Gemini app today

Google brings music to Gemini

Google released its Lyria 3 AI music model for the Gemini app and web interface today. This update moves the technology out of developer-focused environments like Vertex AI and into the hands of general consumers. Users can now generate short musical tracks directly within the Gemini chatbot interface.

The new feature appears as a "Create music" option in the Gemini app and web UI. Users can trigger the model by typing a descriptive prompt or uploading an image to set a specific mood. The system processes these inputs and produces a 30-second audio file in a matter of seconds.

Lyria 3 represents the latest iteration of Google DeepMind’s audio generation research. While previous versions required manual lyric inputs, this model can generate its own lyrics based on vague user descriptions. These 30-second clips function primarily as jingles or background tracks for social media content.

How to generate AI tracks

The interface allows for a high degree of customization through simple text commands. Users can specify genres, instruments, vocal styles, and even the emotional intent of the track. If a user uploads a photo of a sunset, the model attempts to translate those visual cues into a corresponding musical "vibe."

Every music generation task also produces a digital album cover. Google uses its Nano Banana model to create these square images, which accompany the audio file in the Gemini chat window. This creates a complete package for users looking to share their creations on other platforms.

Gemini also includes a library of pre-loaded AI tracks. Users can select these existing songs and use the remix tool to change the tempo, instrumentation, or vocal delivery. This provides a starting point for those who may not want to write a detailed prompt from scratch.

Google is integrating these tools with its broader creative ecosystem. Lyria 3 will power the Dream Track toolkit for YouTube Shorts. This integration allows creators to pair AI-generated music with Veo, Google’s AI video generation model, to automate the production of short-form content.

Prompting and musical capabilities

Google provided several examples to demonstrate the range of the Lyria 3 model. The model handles complex requests involving specific historical genres and technical recording details. It can simulate everything from 1970s soul to modern electronic subgenres.

One example prompt requested a "quintessential 1970s Motown soul" track with a "warm bassline" and "vintage organ harmonic bed." The model responded by generating a track featuring a "three-piece brass section" and "gospel-tinged male tenor lead." This suggests the model understands specific musical arrangements and era-appropriate production techniques.

The model also supports niche formats like a cappella sea shanties. In a test case, Google prompted the model for a "robust male choir" with "synchronized foot-stomps" and "natural room reverb." The resulting audio simulated the acoustic environment of a wooden ship’s deck using only vocal textures and rhythmic percussion.

Users can also request modern, mood-based tracks. One prompt asked for a "wistful and airy" drum and bass rhythm with "rainy city vibes." The model utilized "soft, breathy female vocals" and "deep, warm bass swells" to meet the request. These examples show the model's ability to interpret subjective emotional language into technical audio parameters.

Watermarks and copyright protection

Google is addressing the legal and ethical concerns of AI music by embedding SynthID into every track. SynthID is a digital watermark that remains audible to computers but is imperceptible to the human ear. This technology allows users to verify if a piece of audio originated from Google’s AI by re-uploading it to Gemini.

The SynthID tag persists even if the audio is compressed or modified. This helps streaming services and social media platforms identify AI-generated content in their libraries. Google previously deployed similar watermarking tech for its Imagen and Veo models.

Google also implemented safeguards to prevent the direct imitation of famous musicians. If a user includes a specific artist’s name in a prompt, Gemini will not attempt to copy that artist’s unique voice or signature sound. The system instead treats the name as "broad creative inspiration" for a general genre or style.

The company admits this filtering process is not perfect. If a generated track sounds too similar to an existing artist's copyrighted work, Google encourages users to report the content. This feedback loop helps the company refine its training data and safety filters over time.

Availability and subscription tiers

The rollout of Lyria 3 begins today on the web version of Gemini. Mobile users on Android and iOS should see the "Create music" option appear within the next few days. The initial launch covers a wide geographical area and supports eight primary languages.

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

Google plans to add more languages to the platform in the coming months. While the feature is available to all Gemini users, those with AI Pro and AI Ultra subscriptions will receive priority access. Paid subscribers will have higher daily usage limits for music generation, though Google has not released the specific numbers for these quotas.

Free users will likely face stricter limits on the number of tracks they can generate per day. This tiered approach follows the pattern Google established for its Gemini Advanced features. The company has not yet announced a standalone pricing model for high-volume music generation.

The landscape of AI music

Google’s entry into mainstream AI music puts it in direct competition with startups like Suno and Udio. Those platforms currently offer longer track durations, sometimes exceeding four minutes. Google’s 30-second limit keeps its tool focused on short-form social media use cases for now.

Streaming services like Spotify have already seen an influx of AI-generated tracks. Some of these "fake" artists have accumulated thousands of listeners. Google’s SynthID is an attempt to bring transparency to this trend by making AI origin easily detectable.

The Lyria 3 model is part of a larger trend where tech giants integrate generative media directly into their productivity and communication apps. By placing these tools inside Gemini, Google is betting that users want to create music as easily as they generate text or images. The success of the tool will depend on how creators balance AI convenience with the demand for human-led artistry.