Google has released its Lyria 3 AI music model for public use within the Gemini app, allowing users to generate 30-second songs from text or images. The model produces coherent tracks but is limited to short, mainstream genres and lags behind competitors like Suno and Udio, which offer longer songs and more control. Google has incorporated safety measures like AI watermarking, while its rivals face ongoing legal challenges over copyright claims.
Google DeepMind’s Lyria 3 model is now in beta inside the Gemini app, letting users create fully produced tracks from prompts or photos. The company stated the tool can translate ideas like “*a comical R&B slow jam about a sock finding their match*” into a high-quality track in seconds.
The generated songs are capped at 30 seconds, a limit Google is upfront about, with the goal being shareable moments rather than polished commercial works. Testing showed output is coherent for mainstream genres but struggles with more specific or unusual styles, as acknowledged in its prompting guide.
Rivals Suno and Udio currently lead with longer, structurally complete songs and deeper user controls like prompt strength sliders. Both companies were sued by the Recording Industry Association of America in 2024 for alleged unauthorized training on copyrighted recordings, with Udio settling with Warner Music in November 2025.
Google says it has been “very mindful of copyright and partner agreements” in training Lyria 3, avoiding mimicry of specific artists. All generated tracks include SynthID watermarking, and Gemini can verify if audio was made by its AI, a provenance tool as AI music floods platforms.
The feature is available on desktop for users 18 and older in several languages, with mobile rollout following. It also expands YouTube’s Dream Track feature globally, giving Shorts creators access to AI soundtracks.
