Lyria 3

Overview

Lyria 3 is Google’s generative music model for creating AI-generated audio from text prompts and, in some cases, images. Based on newsletter coverage, it has been rolled out through the Gemini app and later made available via API, shifting it from a consumer-facing creative feature into a developer-accessible media generation capability. The tool has been described as producing original 30-second songs from prompts or images, while Gemini app access also highlighted longer-form generation options such as full-length tracks of roughly three minutes.

For AI Product Managers, Lyria 3 matters because it represents the productization of generative audio across both end-user and platform layers. It is relevant not just as a standalone music model, but as an example of how Google is packaging multimodal generation, safety measures like SynthID watermarking, freemium usage models, and API access for developers. That combination makes it useful for thinking through media feature design, creator tooling, platform monetization, and responsible deployment of generative content.

Key Developments

• 2026-02-19: Google DeepMind rolled out Lyria 3 in global beta via the Gemini app, positioning it as an easy way to generate custom audio. Outputs were noted to include SynthID, Google’s invisible watermark for AI-generated content.
• 2026-02-21: Google AI highlighted Lyria 3 alongside other launches, describing it as a tool that can turn photos or text into dynamic music.
• 2026-04-10: Josh Woodward announced full access to Lyria 3 in Gemini App, with users able to generate up to 5 free full-length (~3 min) tracks per day and unlimited 30-second clips, plus paid upgrades for additional music, image, and video generation.
• 2026-04-11: DeepLearning.AI described Lyria 3 as an AI music generator that transforms text prompts or images into original 30-second songs.
• 2026-05-26: Logan Kilpatrick announced that Lyria 3 became available via API, enabling developers to build products and workflows on top of the model.

Relevance to AI PMs

• Designing generative audio features: Lyria 3 shows how music generation can be packaged across multiple surfaces, from lightweight clip generation to longer-form track creation. AI PMs can use this as a reference for deciding output length, prompt UX, editing workflows, and free-vs-paid feature tiers.
• Evaluating platform readiness: The progression from Gemini app beta to API access is a useful pattern for PMs assessing when a model is mature enough for developer adoption. It highlights questions around documentation, latency, quality consistency, rate limits, and integration pathways.
• Building responsibly with generated media: The mention of SynthID watermarking is especially important for PMs shipping media tools. It points to practical requirements around provenance, disclosure, content policy, and enterprise trust when deploying AI-generated audio.

Related

• Google / Google AI / Google DeepMind: The organizations most directly associated with launching and distributing Lyria 3.
• Gemini App / geminiapp: The initial end-user surface where Lyria 3 was made broadly accessible in beta and then with fuller usage tiers.
• API: A major distribution milestone that expanded Lyria 3 from an app feature into a developer platform capability.
• SynthID: Google’s invisible watermarking system applied to Lyria 3 outputs, relevant for provenance and safety.
• Josh Woodward: Mentioned in connection with expanding Lyria 3 access inside Gemini App.
• Logan Kilpatrick: Announced Lyria 3’s availability via API for developers.
• DeepLearning.AI: Covered the launch and described the model’s prompt-to-music capabilities.