Jeff Chang, Myriam Hamed Torres, and Jason Baldridge from the Google DeepMind team join host Logan Kilpatrick for a deep dive into Lyria 3, Google’s latest music generation model. Their conversation explores the transition from simple audio generation to a model that acts as a collaborative instrument, providing creators with fine-grained control over mood, instrumentation, and vocals. Learn more about the technical challenges of prompt adherence in music, the importance of "vibe" in human evaluations, and the future of layered, iterative music composition.

Chapters: 
0:00 - Intro 
1:00 - Defining music generation models
1:40 - Lyria as a new instrument
3:05 - Connecting language and creative intent
5:08 - Guest backgrounds and musical journeys
7:57 - Demo: Instrumental funk jam
8:29 - Bridging the gap for non-musicians
12:03 - Demo: Exploring lyrics and vocals
15:07 - The magic of iterative co-creation
15:40 - Meeting users across the expertise spectrum
17:01 - Empowering new musical expressions
18:29 - Emotional and communal impact of music
19:51 - Opportunities for developers and community
21:09 - Real-time vs. song generation models
23:23 - Creating experimental sonic landscapes
25:08 - Demo: Capturing unexpectedness and energy
28:33 - Evaluating music through taste and expertise
31:30 - The diligence of music evaluation
31:52 - The future of Lyria and AI-first workflows
35:07 - Articulating creative vision through language


Listen to this podcast: 
Apple Podcasts → https://goo.gle/3Bm7QzQ 
Spotify → https://goo.gle/3ZL3ADl 

Watch more Release Notes → https://goo.gle/4njokfg 
Subscribe to Google for Developers → https://goo.gle/developers 

Speaker: Jeff Chang, Myriam Hamed Torres,  Jason Baldridge, Logan Kilpatrick 
Products Mentioned:  Google AI, Gemini, Genie

Google for Developers

Google DeepMind's Lyria 3 is a new music generation model that accepts text and image inputs to produce high-quality audio, including vocals and instrumentals. The team describes it as a new kind of instrument accessible to both non-musicians and professionals, supporting iterative multi-turn editing and long-context prompting for fine-grained control. Available in the Gemini app and AI Studio, the model is designed to lower the barrier to music creation by letting users describe moods, emotions, or abstract concepts rather than requiring formal musical vocabulary. The team discusses evaluation methodology, future roadmap goals like granular instrument-level editing and real-time generation, and the broader vision of empowering anyone to express themselves musically.

Inside Lyria 3, Google's music generation model