In March 2026, Google DeepMind formally released Lyria 3 — its newest AI music generation model — alongside a public developer preview. Lyria 3 carries forward the audio quality of its two predecessors while making major upgrades in controllability, multimodal input, and the developer toolchain. This guide walks through the full capability system, usage paths, and prompt techniques.
What Is Lyria 3?
Lyria 3 is Google DeepMind’s third-generation music generation model, part of the Gemini model family. It generates high-fidelity, complete musical works — including sung vocals and multi-instrument arrangements — from text descriptions and even from images.
Unlike peers such as Suno and Udio, Lyria 3’s edge is industrial-grade controllability and multimodal input — beyond writing prompts, you can upload an image and have the model compose from what it sees. It is exposed to developers through the Gemini API and Google AI Studio, positioned as integrable, programmable music-generation infrastructure.
Model Tiers: Pro and Clip
Lyria 3 ships in two versions for different scenarios:
| Version | Model ID | Output length | Best for |
|---|---|---|---|
| Lyria 3 Pro | lyria-3-pro-preview | Up to ~3 minutes | Complete songs, professional scoring, high-quality production |
| Lyria 3 Clip | lyria-3-clip-preview | ~30 seconds | Rapid prototyping, BGM loops, social media material |
Pro prioritizes audio quality and structural completeness for finished-music scenarios; Clip is optimized for speed and batch requests, suited to high-frequency, high-throughput calls.
Getting Started
Option 1: Google AI Studio (No Code)
The fastest path, ideal for non-technical users:
- Visit Google AI Studio - Music
- Type a music description in the text box, or upload an image as the creative seed
- Optionally set parameters: BPM, lyric timeline, mood descriptions
- Click generate and wait for Lyria to compose
- Audition and export — professional-grade audio file export is supported
Option 2: Gemini API (Developer Integration)
For developers, Lyria 3 exposes a complete programming interface through the Gemini API:
# Python example: calling Lyria 3 via the Gemini API
import google.generativeai as genai
model = genai.GenerativeModel('models/lyria-3-pro-preview')
response = model.generate_content("A soothing piano piece with soft strings, slow tempo, cinematic atmosphere")
response.audio.export('output.wav')Full documentation lives in the Gemini API music generation docs.
The Core Capabilities in Detail
1. Text to Music
Natural-language generation is the core skill. You can specify:
- Genre and style: EDM, classical, jazz, ambient, 8-bit, lo-fi and more
- Mood and atmosphere: cheerful, melancholic, calm, tense, dreamlike
- Instrumentation: piano, synthesizers, acoustic guitar, string ensembles, electronic drums
- Tempo and rhythm: allegro, adagio, 120 BPM, swing feel
- Structure and arrangement: “start with solo piano, strings enter, build to a climactic chorus”
- Soundscape: rain, city nightscape, spacious reverb, an underwater feel
- Production texture: hi-fi production, clean mix, vintage recording, demo feel
2. Image to Music
This is Lyria 3’s signature advantage — upload an image and Lyria generates music matching its mood, color, and content. A sunset-beach photo yields a soothing guitar piece; a city nightscape might produce an electronic ambient track. For video scoring, the capability is immediately practical.
3. Tempo Control and Time-Aligned Lyrics
Lyria 3 introduces precise tempo conditioning — specify BPM exactly — and supports time-aligned lyrics: declare in the prompt where lyrics begin and end within the song structure, and the model places vocals precisely at the chorus.
4. Multilingual Vocals
Lyria 3 sings in multiple languages — English, Chinese, Japanese among them — across styles from pop and funk to Motown.
5. Digital Watermarking
Every audio output carries a digital watermark, keeping content traceable and verifiable — a meaningful step in Google’s responsible-AI practice.
Prompt Techniques: From Beginner to Pro
Prompt quality is the main lever on results. Three ascending examples:
Basic prompt (ordinary result)
Ambient music with synthesizersResult: a generic ambient fragment with unremarkable mood and structure.
Refined prompt (clear improvement)
A calm and dreamy ambient soundscape featuring layered synthesizers and soft, evolving pads. Slow tempo with a spacious reverb. Starts with a simple synth melody, then adds layers of atmospheric pads.Result: focused music, a richer soundscape, clear progressive layering.
Advanced prompt (professional output)
An energetic electronic dance track with a fast tempo and a driving beat, featuring prominent synthesizers and electronic drums. High-quality production. Structure: intro → build-up → drop → breakdown → final drop → outro.Result: a crisp, punchy 30-second instrumental with distinct structure.
Real-World Applications
- Video scoring: AI Studio ships a demo app — upload a video, Gemini analyzes the footage and drafts the prompt, and Lyria generates matching background music
- Game audio/BGM: use Lyria 3 Clip to mass-produce loop tracks
- Composition assistance: generate inspiration fragments, then polish by hand
- Podcast/livestream BGM: scene-matched background music on demand
- Social short video: AI soundtracks inside YouTube Shorts via Dream Track
Competitive Positioning
Against Suno v5 and Udio, Lyria 3’s differentiation:
- Multimodal input: Suno and Udio remain primarily text-driven; Lyria takes both text and images
- Developer toolchain: enterprise-grade integration through the Gemini API and Vertex AI
- Controllability: BPM control and time-aligned lyrics are finer-grained than the competition’s
- Ecosystem integration: deep linkage with Gemini, Google Vids, and YouTube
To be fair, on pure end-to-end song quality Suno v5 remains the industry benchmark. But Lyria 3 offers a different proposition — a more professional, more controllable, more developer-integrable music generation infrastructure.
Get Started
Lyria 3 is in public preview and free for all developers to try:
- Try online: Google AI Studio - Music
- API docs: Gemini API Music Generation
- Cookbook and samples: official examples and code
- Product page: Google DeepMind - Lyria
If you’re looking for an AI music solution you can integrate programmatically, Lyria 3 is currently the option to watch.
