Google Unveils Lyria 3: AI Music Studio Integrated Directly Into Gemini

On February 18, 2026, Google DeepMind announced the global rollout of Lyria 3, its most advanced music generation model to date. Moving beyond research previews, Google has made the tool available to users worldwide via the Gemini web interface and app, effectively transforming the chatbot into a comprehensive music production studio.

New Features: Vocals, Video, and Multimodal Input

Lyria 3 significantly expands upon the capabilities of previous experimental versions:

• Multimodal Input: Users are no longer limited to text prompts. The model can analyze uploaded photos or videos to generate a soundtrack that matches the visual rhythm and mood (e.g., scanning a video of a rainy street to produce lo-fi jazz).
• Lyric & Vocal Generation: Unlike earlier iterations, Lyria 3 can write lyrics and generate vocal performances. It currently supports vocals in 8 languages, including English, Spanish, Japanese, Korean, and Hindi (with Arabic available in beta).
• Granular Control: New interface controls allow users to adjust tempo, genre style, and instrumental "density." The model generates high-fidelity 30-second clips, which can be seamlessly extended or looped.

Integration with 'Nano Banana'

To provide a complete creative package, Google has integrated its latest image generation model, internally codenamed "Nano Banana" (part of the Gemini 2.5 Flash Image family). This system automatically analyzes the lyrics and mood of the generated track to create unique, high-quality album artwork for every song.

Safety and Copyright

Google emphasized that Lyria 3 was trained with strict adherence to copyright protection and artist safety.

• Anti-Mimicry Guardrails: The model is designed to reject prompts asking to replicate specific artists. If a user requests a track "in the style of Taylor Swift," the system uses the query for broad creative inspiration only, ensuring it does not clone the artist's voice or signature melodic structures.
• SynthID: All audio outputs are embedded with SynthID, an imperceptible watermark that remains detectable even if the audio is compressed, edited, or mixed, ensuring AI-generated content can always be identified.

Availability

The feature has begun rolling out today to Gemini users (18+) globally. Google positions this release as a direct competitor to services like Suno and Udio, leveraging its deep ecosystem integration to bring advanced music creation tools to a mass audience.