Google Gemini Can Now Generate AI Music with Lyria 3 Model

Google’s Gemini virtual assistant now boasts the ability to generate original music, powered by the company’s latest AI model, Lyria 3. The new feature, currently available in beta, allows users to create 30-second tracks from both text prompts and image uploads.

Lyria 3, described by Google as its “most advanced music generation model yet,” aims to deliver high-fidelity audio complete with instrumentals, vocals, and lyrics. The technology has been developed with input from music producers and musicians, focusing on understanding musicality – encompassing rhythm and arrangement – to create cohesive songs.

The rollout of Lyria 3 arrives amidst a broader conversation about the role of artificial intelligence in the creative industries. While offering new avenues for musical expression, the technology also raises complex questions about copyright and the potential displacement of human artists. Google states it is “mindful of copyright and partner agreements,” and that Lyria 3 is trained only on music that YouTube and Google have the rights to use under their terms of service.

However, concerns remain about the model’s training data. Ed Newton-Rex, a prominent voice in the debate surrounding AI and music, expressed skepticism on X (formerly Twitter), suggesting that the claim of being “mindful of copyright” could be “intentional misdirection.” He urged users to assume the model was trained on copyrighted work without a license unless explicitly stated otherwise.

The launch of Lyria 3 follows Google’s recent release of Project Genie, a creation tool that has already faced criticism for enabling the replication of existing games like The Legend of Zelda, Mario, and Grand Theft Auto. This pattern highlights a growing trend of AI tools potentially infringing on existing intellectual property.

The music industry is actively grappling with the challenges posed by AI-generated content. Apple Music recently demonetized two billion “fraudulent” music streams, amounting to approximately $17 million in royalties, while Deezer has demonetized 85 percent of all AI-generated tracks on its platform, and is now offering an AI-detection tool commercially. Bandcamp has taken a more decisive stance, banning all AI-created tracks outright.

Despite these efforts, distinguishing between human-created and AI-generated music remains difficult. A recent study found that 97 percent of people are unable to tell the difference. This ambiguity is further complicated by the potential for AI to impact the livelihoods of music professionals, with one study predicting a potential loss of nearly a quarter of all work in the sector over the next four years.

Recent incidents have underscored the complexities of AI-generated music. A song that garnered millions of streams in Sweden was banned from the charts due to its AI origins. Simultaneously, questions arose regarding the authenticity of singer Sienna Rose, with suspicions that her music was entirely AI-generated. The case of The Velvet Sundown, an AI ‘band’ that quickly amassed a significant Spotify following before being revealed as a hoax, further illustrates the deceptive potential of the technology.

Leading artists, including Paul McCartney, Kate Bush, and Elton John, have publicly urged UK Prime Minister Keir Starmer to protect creative work in the face of the AI threat. Starmer acknowledged the importance of protecting creativity, stating that his government is currently reviewing responses to a consultation on the issue, aiming to strike a balance between innovation and copyright protection.

The debate extends beyond legal frameworks. Artists like Mac DeMarco and SZA have voiced criticism of AI-generated music, while ABBA’s Björn Ulvaeus has embraced it as a “great tool.” Kehlani recently criticized an AI-generated artist receiving a $3 million record deal, highlighting the perceived unfairness of the situation.

Even established artists are experimenting with AI. Teddy Swims admitted to occasionally using the technology in his music-making process, while Jorja Smith’s record label criticized a song that allegedly cloned her voice using AI.

Google is attempting to address concerns about transparency by embedding all tracks generated in the Gemini app with SynthID, an imperceptible watermark for identifying AI-generated content. The company is also expanding its verification capabilities to include audio, allowing users to upload files and determine if they were created using Google AI.

Lyria 3 is currently available globally in English, Japanese, Korean, Hindi, Spanish, Portuguese, German, and French for users over the age of 18. Access to the feature is automatic for users with access to the Gemini app, with increased limits available for subscribers. The rollout is being managed through Workspace Admin controls for Google Workspace customers.