🇺🇸 United States · Google Lyria — AI Music Generation Model
Status: 🟩 COMPLETE 🟦 LIVING Section: 10 — AI and LLMs
| Vendor | Google DeepMind |
| Country/origin | 🇺🇸 United States / 🇬🇧 United Kingdom (Google DeepMind) |
| Recommended for AUS? | ✅ Yes — Google standard privacy; part of Google DeepMind |
| Privacy summary | Google’s standard data handling; inputs may be used to improve models unless opted out; enterprise controls for Workspace users |
| Free tier | Limited — available through certain Google products |
| Paid tiers | Via Google One AI Premium (~$30 AUD/month) and YouTube Premium features |
| First released | Lyria research published 2023; consumer integration 2024–2025 |
| Last reviewed | June 2026 |
| Official site | https://deepmind.google/discover/blog/lyria-google-deepminds-music-generation-model |
What it is
Lyria is Google DeepMind’s AI music generation model — a model trained to create high-quality audio music from text prompts and capable of generating music in a wide range of styles.
Lyria is not a standalone product you can use directly. It’s an underlying model that Google integrates into its various products and research tools. It appears in:
- YouTube Dream Track: A feature for YouTube Shorts creators that generates background music from a text description, in the style of various licensed artists (with those artists’ permission and revenue sharing)
- Music AI Tools (collaboration with YouTube): Experimental tools for musicians to generate and transform audio
- SynthID integration: All Lyria-generated audio is watermarked with Google’s SynthID system for AI content identification
How it works
Lyria uses a diffusion-based approach (similar to image generation diffusion models, but applied to audio spectrograms — the visual representation of sound as frequencies over time). Key capabilities:
- Text-to-music: Generate instrumental or voiced music from a text prompt
- Music continuation: Continue a piece of music in the same style
- Style transformation: Take an existing audio clip and transform it into a different musical style
- Instrument-level control: Some research features allow specifying which instruments should be present
How to access (Australian users)
Access to Lyria in mid-2026 is through Google’s partner integrations:
- YouTube Dream Track: Available to eligible YouTube Shorts creators — in the YouTube app (mobile), tap + → Create a Short → Soundtrack → Dream Track → describe the music you want
- Music AI Tools (limited availability): Applied through the AI Test Kitchen or specific creator programs
- Research access: Academic partners can apply for Lyria API access
For most Australian users, YouTube Dream Track is the primary access point — and it requires being a YouTube creator with access to Shorts.
How it compares to Suno and Udio
| Aspect | Lyria | Suno | Udio |
|---|---|---|---|
| Access | Very limited | Open | Open |
| Integration | YouTube Shorts native | Standalone | Standalone |
| Full songs | Research only | ✅ | ✅ (via extend) |
| Instrumental | ✅ | ✅ | ✅ |
| Commercial rights | Via YouTube artist deals | Paid plan | Paid plan |
| SynthID watermarking | ✅ Always | ❌ | ❌ |
| Free access | Very limited | ✅ 10 songs/day | ✅ 100/month |
Lyria’s main advantages: AI content watermarking (SynthID), artist collaboration model (licensed likeness with revenue share), YouTube integration. Main limitation: Very restricted access compared to Suno and Udio.
The artist licensing model
One distinctive aspect of Lyria’s YouTube Dream Track implementation: rather than training on artists’ music without permission (the approach that led to lawsuits against Suno and Udio), Google partnered with specific artists who agreed to license their musical style for AI generation. These artists receive revenue sharing when their “style” is selected.
Participating artists in early rollouts (2024) included Alec Benjamin, Charlie Puth, Charli XCX, Demi Lovato, John Legend, Sia, T-Pain, and others.
This model addresses the copyright concerns around AI music generation — though critics note that the “style” of music cannot be copyrighted, making the partnership a commercial arrangement rather than a legal necessity.
SynthID audio watermarking
All audio generated by Lyria includes Google’s SynthID audio watermark — an imperceptible signal embedded in the waveform that can be detected by Google’s tools. This is part of Google’s AI content provenance strategy. See watermarking-ai-content for more detail.
Gotchas
- Lyria is not a consumer product you can directly use — it’s an underlying model accessed through specific Google products with narrow availability.
- YouTube Dream Track is geographically limited. Rollout has been gradual; not all regions have access.
- For music creation, Suno and Udio are more accessible and produce full songs with vocal tracks. Lyria’s strength is in specific YouTube integration use cases.
- The research/experimental tools are invitation-only. Applying to use Lyria directly as a musician requires going through Google’s creator programs.
See also
- music-generation — full AI music generation overview; Suno and Udio are more accessible
- suno — most accessible AI music tool
- udio — highest-quality vocals; alternative to Suno
- watermarking-ai-content — SynthID watermarking
- google-deepmind — the lab behind Lyria
Sources
- Google DeepMind Lyria blog post (2023): deepmind.google/discover/blog/lyria
- YouTube Dream Track announcement and documentation (2024)
- MusicLM technical paper (precursor to Lyria): García et al. (2023)
- Music Business Worldwide — artist partnerships for YouTube Dream Track (2024)
- Google SynthID audio watermarking paper (2023)