Google has quietly expanded its Gemini AI beyond text and images into music creation. With the debut of Lyria 3, a specialized model developed by DeepMind, users can now prompt the system to generate original 30-second musical compositions—complete with lyrics—by describing a mood, uploading an image, or referencing a video.
The feature, rolling out globally in beta, isn’t just about convenience. It’s designed with creative control in mind: users can tweak tempo, genre, vocal style, and even the emotional tone of the output. For example, a prompt like a dreamy synthwave track about lost keys, slow tempo, no vocals* might yield a short, atmospheric piece tailored to that exact request.
But there’s a catch. Every AI-generated track carries Google’s SynthID watermark, an invisible digital fingerprint that identifies the music as machine-made. The system can also detect whether existing audio files were created using Lyria 3, adding a layer of transparency rarely seen in AI tools.
A Tool for Creators, Not Copiers
Google has made it clear: Lyria 3 is built for original creation, not imitation. The model avoids replicating specific artists’ styles, focusing instead on generating fresh compositions. That said, the technology could still raise questions about copyright and artistic ownership—though Google hasn’t detailed how it plans to handle commercial use or distribution of the generated tracks.
Access isn’t limited to English speakers. Lyria 3 supports eight languages, including Spanish, Hindi, Japanese, and Portuguese, with the requirement that users be at least 18 years old. Paid Gemini subscribers will enjoy higher usage limits, though the company hasn’t specified exact thresholds.
How It Works in Practice
The process is straightforward. Users start with a text prompt—something like a fast-paced electronic track for a cyberpunk short film*—or upload an image (e.g., a sunset) or video (e.g., a montage of city lights). Gemini then generates a 30-second song with matching lyrics, if applicable. The interface allows adjustments to tempo, instrumentation, and vocal presence, giving creators fine-grained control over the final product.
For now, the feature remains in beta, meaning bugs and limitations may still exist. But if the execution lives up to the promise, Lyria 3 could redefine how non-musicians and hobbyists approach songwriting—all while keeping the AI’s hand visible through SynthID.
- Model: Lyria 3 (DeepMind-developed)
- Output: 30-second original songs with lyrics (optional)
- Inputs: Text prompts, images, or videos
- Customization: Genre, mood, tempo, vocals
- Watermark: SynthID for AI detection
- Languages: English, German, Spanish, French, Hindi, Japanese, Korean, Portuguese
- Age restriction: 18+
- Availability: Global beta in Gemini app
- Subscription perk: Higher usage limits for paid users
While Lyria 3 won’t replace professional music production, it lowers the barrier for experimentation. Whether that translates into a flood of AI-generated hits—or just more background music for creators—remains to be seen.
