Google is stepping up its investment in the generative AI content creation space.
On Wednesday, March 29, Google announced the launch of its next-generation AI music generation model, Lyria 3.5, which is now integrated into Flow Music, an AI music creation tool under Google Labs. Compared to its predecessors, the new model delivers major upgrades in musicality, lyric generation, AI vocals, and creative control, enabling users to produce songs that are more natural, structurally complete, and emotionally rich.
This update signals Google's push to strengthen its text-to-song generation capabilities amid growing competition from products like OpenAI's, Suno, and Udio. The goal is to position Flow Music as a one-stop AI music platform, covering everything from creation and editing to final song generation.
Four Key Upgrades in Lyria 3.5
According to Google, Lyria 3.5 focuses on four key areas of improvement.
The first is musical expression. Google states that the new model can generate more complex melodic structures, making the music more natural and fluid overall. Song arrangements now follow real-world music creation logic more closely, rather than being simple repetitive loops.
The second is enhanced lyric generation capability. Lyria 3.5 has improved prompt adherence, allowing it to understand user requests more accurately. It produces songs with better structure and higher-quality lyrics, showing significant improvement in paragraph organization and overall coherence.
The third, and most significant, upgrade is in AI vocals. Google says the new model generates singing voices that are more natural and emotionally nuanced, with improved pronunciation accuracy. This makes the AI-generated vocals sound much closer to a real human singer.
Additionally, users can now exert more precise control over song tempo and duration, which further enhances creative freedom.
The fourth area is enhanced creative control. Beyond improving generation quality, Google has placed a strong emphasis on creative control. Google says Flow Music aims to make AI more of a creative partner rather than just a one-time music generation tool.
Users can now continuously modify songs using natural language prompts, iterating on melody, lyrics, style, tempo, and overall mood without needing to regenerate the entire song from scratch each time. Google hopes Lyria 3.5 will help users "create songs they truly love," not just generate a demo.
Continuous Iteration of the Lyria Series
Lyria is a series of AI music generation models developed by Google DeepMind and is a key component of Google's recent push into AI content generation.
In February of this year, Google launched Lyria 3. By the end of March, it released Lyria 3 Pro, which extended the maximum song generation length from around 30 seconds to 3 minutes. This version also allowed users to define complete song structures like intros, verses, choruses, and bridges, making the AI-generated content more aligned with professional music production workflows.
With the release of Lyria 3.5, Google did not announce new benchmark scores. Instead, the upgrade focused on user experience aspects like vocal quality, lyric creation, and creative control. Industry observers believe this indicates Google aims to improve the usability of AI-generated music rather than just pursuing raw model performance metrics.
Intensifying Competition in AI Music
In recent years, AI music has become a significant competitive frontier in the generative AI landscape.
Besides Google, startups like Suno and Udio have risen rapidly. OpenAI is also consistently exploring multimodal content generation capabilities. Various companies are now competing fiercely in areas such as song creation, audio editing, and vocal generation.
As an integral part of Google Labs' creative tool suite, Flow Music is working alongside the Flow video generation platform to build Google's broader AI content creation ecosystem. With the launch of Lyria 3.5, Google aims to integrate AI capabilities more deeply into the entire music creation process, from initial ideas and lyric writing to song generation and editing, further solidifying its competitiveness in the generative creation space.
Comments