Lyria 3 by Google DeepMind: Google Finally Enters the AI Music Game—But Is the Giant Too Late?

2026-02-22 | ProductHunt | Official Website

Open the Gemini App, tap "Create Music," enter a sentence or a photo, and 30 seconds later, you have an original song with vocals, lyrics, and cover art. That’s the magic of Lyria 3.

30-Second Quick Judgment

What is it?: Google has packed its most powerful music generation model, Lyria 3, into the Gemini App. Anyone can generate a 30-second song with vocals using just text or images, completely for free.

Is it worth your attention?: Yes, but don't get over-excited just yet. If you just need a fun background track for a short video or a social media post, this is perfect. However, if you're looking to create full-length songs, Suno and Udio are still miles ahead. Google's entry signals that the AI music race has officially entered the "clash of the titans" phase—that's the real signal to watch.

Three Questions for Me

Is it relevant to me?

Who is the target user?:

• Hundreds of millions of existing Gemini users (18+)
• Short-video and social media content creators
• Casual users who want to play with music but have no musical background
• YouTube Shorts creators (via Dream Track)

Am I the target? If any of these apply, then yes:

• You often need background music for videos but don't want to pay for licenses.
• You want to send a funny birthday song to a friend but can't write music.
• You're a YouTube/TikTok creator needing quick soundtracks.
• You simply want to see what AI is capable of.

When would I use it?:

• You took a travel photo → Let Lyria 3 create a 30-second track that matches the vibe.
• A short video needs BGM → Generate it in one sentence, for free, with no copyright worries.
• Making memes → Input "A rock song about my cat thinking he's a lion" → Laugh 30 seconds later.
• Not suitable for: Full-length songs, professional production, or commercial releases.

Is it useful to me?

ROI Judgment: For content creators, this is a great "freebie" tool. It takes 5 minutes to learn and costs $0. While 30 seconds is short, it's plenty for short-form content. The ROI is high. However, if you're a musician or need professional output, the ROI is near zero—30 seconds of "corny" lyrics won't help you much.

Is it fun to use?

The "Wow" Factor:

• Zero-to-Song Speed: For someone who has never made music, hearing "their song" in 30 seconds is a genuine "wow" moment.
• Image-to-Music: Turning a photo into a song is something Suno/Udio don't offer yet, making it Lyria 3's most unique feature.
• Ecosystem Integration: If you're already using Gemini, there's no need to sign up for a new service.

A "Wow" Moment:

"I tried Gemini's new Lyria 3 in-app AI song generator -- and it turned my to-do list into a punk rock anthem" -- Tom's Guide Review

Real User Feedback:

Positive: "I've tested dozens of AI generated music tools, but Lyria 3 is different because it's so much easier to use." Neutral: "It works, it's fun, and it may impress anyone who has never used Suno or Udio. For those who have, it's not going to replace their workflows anytime soon." -- Decrypt Review Negative: "The last thing the world needs right now." -- ProductHunt commenter Sharp: "Gemini will now generate musical slop for users" -- The Register

For Independent Developers

Tech Stack

• Core Architecture: Multimodal Diffusion Transformer (MMDiT)
• Audio Generation: Latent Diffusion + Rectified Flow (RF)
• Text Understanding: LLM-type language encoder
• Audio Output: 48kHz stereo, supports WAV/MP3/FLAC
• Real-time Streaming: Block autoregression based on MusicLM, WebSocket bi-directional connection, 2-second chunks
• Watermarking: SynthID imperceptible watermarking
• Infrastructure: Google Cloud / Vertex AI

Core Implementation

At its heart, Lyria 3 is a multimodal diffusion model. It encodes text, images, and video into a "shared embedding space" and then uses a diffusion process to denoise Gaussian noise into a 48kHz audio waveform. Essentially, it applies the Stable Diffusion image generation logic to the audio domain.

One of the most interesting aspects is Lyria RealTime. It uses WebSockets to maintain a bi-directional connection, generating audio in 2-second chunks while maintaining "rhythm" based on previous context and adjusting style based on user input. Google even released a VST plugin called "The Infinite Crate" for use in DAWs.

Open Source Status

• Is it open source?: No, it is proprietary to Google.
• Similar open-source projects: Meta's MusicGen, Stability AI's Stable Audio.
• API Status: The Lyria RealTime API is available (streaming, instrumental only), and Lyria 2 is available via Vertex AI (lyria-002). The full Lyria 3 API is not yet public.
• Difficulty of DIY: Extremely high. Music generation is much harder than text or images—it requires handling melody, harmony, rhythm, and timbre simultaneously while maintaining long-range consistency. Without Google-scale compute and data, it's nearly impossible for independent devs to match this quality.

Business Model

• Monetization: Not directly monetized; it's part of the Gemini ecosystem designed to increase user stickiness and the value of Google AI subscriptions.
• Indirect Monetization: Drives Google AI Plus/Pro/Ultra subscriptions (for higher generation limits).
• User Base: Global Gemini user base (hundreds of millions).

Giant Risk

This product is the definition of "giant risk." For Suno and Udio, Google's entry is a serious threat—Lyria 3 is free, integrated into Gemini, and backed by the YouTube ecosystem. However, the current 30-second limit means Suno and Udio still have a window of opportunity. The big question is: how fast can Google catch up?

For Product Managers

Pain Point Analysis

• What problem does it solve?: It allows people with zero musical background to "create" music.
• How painful is the pain point?: Medium-low. While content creators have a high-frequency need for BGM, existing solutions (stock libraries, Suno/Udio) are already quite good. Lyria 3's core edge is being "free" and "built-in."

User Persona

• Core User: Content creators among existing Gemini users.
• Secondary User: Casual users curious about AI music.
• Non-target User: Professional musicians or creators needing full-length songs.

Feature Breakdown

Competitive Comparison

Key Takeaways

1. "Integration Advantage": Don't build a standalone app; embed it into a platform with millions of users. This is more effective than any marketing.
2. Multimodal Thinking: Adding image-to-music as a new input dimension significantly lowers the creative barrier.
3. "For Fun" Positioning: By not chasing professional-grade output initially, they avoid the heat of copyright battles and capture user mindshare as a "fun toy."

For Tech Bloggers

Founder/Team Story

This isn't a startup story; it's a decade-long product line inside Google:

• 2016: Douglas Eck founds the Magenta Project at Google Brain to explore AI and music.
• 2023: Google DeepMind releases MusicLM, showing text-to-music capabilities.
• 2024: Lyria 1 (Dream Track on YouTube) is released, tested with famous artists.
• 2025: Lyria 2 API opens, and Lyria RealTime is introduced.
• Feb 2026: Lyria 3 integrates into the Gemini App for all users.

Core members include Andrea Agostinelli (lead developer across Lyria projects) and Myriam Hamed Torres (Product Manager). The spiritual leader is Douglas Eck, who is frequently credited for his "insightful guidance."

Points of Contention

1. "Too Little Too Late": Decrypt's review title says it all: "We Tried It, and It's Too Little Too Late." What took Google ten years, Suno achieved in two. Is the giant's R&D advantage failing in the fast-paced AI era?

2. "Musical Slop" vs. "Creative Democracy": While The Register calls it "musical slop," Google frames it as "creative expression." Is this lowering the barrier or just devaluing music?

3. "Obsolescence by Trivialization": The Next Web suggests Lyria 3 isn't meant to replace musicians, but to make platforms no longer need them. When anyone can generate a "good enough" song in 30 seconds, where is the value of a pro?

4. The "Original Sin" of Training Data: Billboard reported in 2024 that DeepMind "trained first and asked for permission later." Google now says they are "mindful of copyright," but won't disclose training details—the same issue Suno/Udio face, just with better PR.

Buzz Data

• PH Ranking: 3 votes—almost no heat. Likely because it's a feature update rather than a standalone product.
• Media Coverage: Explosive. TechCrunch, Billboard, Tom's Guide, and Music Business Worldwide all covered it immediately.
• Social Media: High volume of shared songs on X (Twitter); Google DeepMind's official announcement sparked wide discussion.

Content Suggestions

• The Narrative Angle: "Google's Late-to-the-Party Music Ambition"—the 10-year story from Magenta to Lyria 3.
• Viral Opportunity: Create a "Songs for every job" series using Lyria 3; it's easy to produce and highly shareable.
• The Comparison: Run the same prompt through Lyria 3, Suno, and Udio for a head-to-head showdown.

For Early Adopters

Pricing Analysis

Quick Start Guide

• Time to start: 5 minutes
• Learning curve: Very low
• Steps:
  1. Open the Gemini App or web version.
  2. Click the "Create Music" button in the toolbar.
  3. Enter a description (e.g., "A lo-fi slow jam about working late, with a hint of humor") or upload a photo.
  4. Wait a few seconds and play the song.
  5. Not happy? Reply with changes ("Make the tempo faster" or "Change to a female voice").

Prompt Tips:

• Specify Style + Mood + Specific Content (e.g., "A comical R&B slow jam about a sock finding its match").
• Combining photos with text works best.

Pitfalls and Gripes

1. 30 seconds is too short: This is the biggest complaint. If you want a full song, use Suno.
2. Corny lyrics: Auto-generated lyrics are often described as "cheesy" or "strange." Don't expect poetry.
3. Niche genres struggle: Pop, R&B, and light Hip-Hop are great; experimental electronic or traditional folk are not.
4. No API: You can't call Lyria 3 in your own app yet; only the instrumental Lyria RealTime is available.
5. Daily limits: Free users have a cap, though Google hasn't disclosed the exact number.

Safety and Privacy

• Data Storage: Cloud-based (Google servers).
• AI Watermark: All outputs contain SynthID, detectable as AI-generated.
• Privacy: Follows standard Google privacy policies.
• Artist Protection: Filters are in place to prevent imitating specific artists, though Google admits they "might not be foolproof."

Alternatives

For Investors

Market Analysis

• AI Music Software Market: $1.18B in 2026 → $7.29B in 2036, CAGR 20.1%.
• Broader AI Music Market: $5.2B in 2024 → $60.4B in 2034, CAGR 27.8%.
• Drivers: Content explosion (Shorts/Podcasts/Gaming), high cost of licensed music, maturing AI tech.

Competitive Landscape

Timing Analysis

• Why now?: 2024-2025 saw AI music move from "demo" to "product." Suno and Udio proved PMF. 2026 is when giants enter to harvest the market.
• Tech Maturity: Diffusion models for audio are now mature; 48kHz high-fidelity output is the new standard.
• Market Readiness: User acceptance of AIGC has skyrocketed, though copyright battles remain the biggest uncertainty.

Team Background

• Parent Company: Google / Alphabet (Market Cap ~$2T).
• R&D Team: Google DeepMind, world-class AI research.
• History: 2016 Magenta → 2023 MusicLM → 2024 Lyria 1 → 2026 Lyria 3.

Funding Status

• Lyria 3 is an internal Google product; no independent funding needed.
• Comparison: Suno raised $125M in 2024; Udio was acquired by Universal Music.
• Signal: Tech giants entering means the market is validated, but the window for independent startups is narrowing.

Conclusion

One-sentence Judgment: Google has built a "good enough free toy" in Lyria 3. It won't replace Suno, but it will likely be the first AI music experience for millions—and that "market education" is more valuable than the product itself.

Resource Links

2026-02-22 | Trend-Tracker v7.3