Gemini Can Now Write You a Song

Google has expanded its multimodal capabilities with the launch of Lyra 3, a new AI music generation model now available directly within the Gemini app. This latest iteration allows users to create 30-second audio clips based on text, image, or video prompts, marking a significant shift from cloud-only access to consumer-facing integration. The tool also incorporates Google's SynthID audio watermarking to ensure AI-generated content remains identifiable.

Key Points

• Lyra 3 generates 30-second audio clips from text, images, or video, supporting lyrics in eight languages including Hindi, French, and German.
• The feature is integrated into the Gemini app and YouTube’s Dream Track, positioning the tool as a social-first creative outlet rather than a professional production suite.
• Google is prioritizing accessibility, moving Lyra from limited Vertex AI cloud access to broad user availability.
• The model is designed for short-form content, such as background music for YouTube Shorts, and does not support the generation of full-length songs.

Expanding the Multimodal Ecosystem

The release of Lyra 3 represents Google’s broader strategy to weave AI into every facet of its digital ecosystem. While competitors like Suno focus on high-fidelity, long-form song generation, Google has positioned its latest model as a lightweight social feature. By allowing users to pair custom, AI-generated cover art with musical snippets, the company aims to simplify personal expression for casual creators.

Industry observers have noted the technical achievement behind the tool, particularly its ability to maintain sync between visual inputs and audio outputs. Chaien Xhiao noted the difficulty of this integration, stating: "Video to audio alignment is the real flex here. Generating lyrics and vocals that actually sync with visual cues in real time is a massive multimodal serving challenge."

Strategic Walled Gardens

The launch comes amid broader industry friction regarding how AI labs restrict access to their models. Recently, Anthropic clarified its terms of service following user backlash over the use of authentication tokens in third-party "agent" tools like OpenClaude. Anthropic’s reach head, Shahipard, attempted to mitigate concerns by emphasizing that the company intends to distinguish between personal tinkering and unauthorized commercial use of its API.

However, the incident highlighted the growing trend of "walled gardens" among major AI providers. As firms like Google, OpenAI, and Anthropic tighten their terms, developers and power users are finding less flexibility in how they integrate these models into independent applications. Colin Darling, a vocal observer of these shifts, noted that while Anthropic drew criticism, the company is merely aligning with the restrictive policies already standard at Google and OpenAI.

Meta’s Wearable Ambitions

Beyond software, the hardware race continues to heat up. Reports indicate that Meta has revived its smartwatch project under the internal code name Malibu 2. This device, expected to launch later this year, is slated to feature health-tracking capabilities and an integrated Meta AI assistant. This move signals Meta’s commitment to a multi-device strategy that includes their Orion prototype glasses and Ray-Ban smart eyewear.

"The goal of these tracks isn't to create a musical masterpiece, but rather to give you a fun, unique way to express yourself." — Google official statement regarding Lyra 3

As Meta, Apple, and Google continue to refine their respective AI stacks, the focus is shifting toward "concentrated bets." Meta has reportedly delayed other AR projects to focus resources on the new smartwatch and upcoming iterations of their smart glasses, aiming to ensure their AI assistants remain accessible even when users are away from their screens.

Looking Ahead

The market remains flooded with new models and benchmark claims, which industry experts advise approaching with caution. As Lindy founder Flo Crell noted, performance on standardized benchmarks often fails to translate to real-world agentic behavior. Moving forward, the true test for Lyra 3 and similar tools will be whether they can move beyond the "experiment" phase and provide consistent, practical value to the average user, or if they will remain relegated to niche social novelty. The industry now awaits the next evolution of these models to see if they can bridge the gap between impressive demonstrations and utility.