Google Lyria 3.5 is moving from a specialist creation tool into mainstream Gemini distribution. On September 4, Google made the music-generation model available to all Gemini users globally on web and mobile while also opening Lyria 3.5 to developers through the Gemini API.

 

The expansion matters because Google is no longer treating advanced music generation as a separate experiment. Lyria now sits inside the same consumer and developer ecosystem as Gemini, giving casual users, creators and software teams a direct path from a text or image prompt to a finished song.

 

Related Research

 

Google Lyria 3.5 Moves Into Gemini

Google describes Lyria 3.5 as its best-sounding music-generation model, with improvements in vocal expression, musical arrangement and overall fidelity. In the Gemini app, users can describe or select a genre, choose vocal or instrumental output, use starter templates and request either shorter or longer tracks.

 

The consumer rollout is broader than the model's earlier placement in Flow Music. Google says Lyria 3.5 is now available globally in Gemini, while artists and AI creators can still use it through Flow Music and developers can access the model through Google AI Studio and the Gemini API.

 

Google DeepMind's Lyria page says the model can generate cohesive tracks up to three minutes long. It can also work from images, turning visual material into a musical direction rather than limiting generation to text prompts alone.

 

The Gemini version gives users several direct controls:

  • Genre selection or free-form genre descriptions
  • Vocal or instrumental track generation
  • Templates for uses such as backing tracks, jingles and personal songs
  • Shorter or longer track lengths
  • Image-guided music creation through Lyria's multimodal capabilities

 

What Developers Get in the Gemini API

The developer release turns Lyria 3.5 into a programmable music service rather than only a consumer feature. Google's API documentation says the model generates 44.1 kHz stereo audio and can produce songs with vocals, timed lyrics and full instrumental arrangements.

 

Google lists two main music-generation models. Lyria 3 Clip, identified as lyria-3-clip-preview, always produces 30-second clips for loops and previews. Lyria 3.5, identified as lyria-3.5, is designed for full songs with verses, choruses and bridges lasting a couple of minutes.

 

Developers can provide text prompts or as many as 10 images alongside a prompt. They can also supply their own lyrics with structural markers such as verse, chorus and bridge tags, and use timestamps to specify when instruments or sections should enter.

 

That makes the API relevant to products beyond a simple song generator. Video tools can create backing tracks from visual references, marketing systems can generate short brand audio, and creative applications can expose structured controls without building a music model from scratch.

 

Google's Guardrails for AI Music

Google is shipping Lyria 3.5 with explicit limits around imitation and copyrighted material. Its developer documentation says safety filters block prompts requesting a specific artist's voice and requests to generate copyrighted lyrics.

 

Every generated track also carries Google's SynthID audio watermark. DeepMind describes the watermark as imperceptible to listeners while still allowing systems to identify music that was created or edited using Google's AI tools.

 

The current model is not an unrestricted digital audio workstation. Google says music generation is single-turn, meaning users cannot repeatedly edit the same generated clip through a chain of follow-up prompts. Results can also vary between calls even when the prompt is unchanged.

 

Those limitations are commercially important as AI music systems face growing scrutiny from artists, labels and courts. The Next Web noted that Lyria's restrictions on artist imitation and copyrighted lyrics arrive amid active copyright disputes over music-generation systems, increasing pressure on developers to show how their models control obvious infringement risks.

 

The Distribution Shift for Generative Music

Lyria 3.5's biggest change is distribution. A high-fidelity music model that once required users to enter a dedicated creative environment is now embedded in Gemini and exposed as an API that other software can call directly.

 

That puts music generation closer to the workflows where people already use AI for writing, images, video and coding. A creator can move from an idea to lyrics, artwork, video and now a soundtrack inside a connected Google toolchain instead of moving between unrelated services.

 

For developers, the API may be even more consequential. Google is effectively offering music generation as infrastructure, with predictable model IDs, multimodal inputs and structured output options that can be incorporated into consumer apps or enterprise creative software.

 

The competitive question now shifts from whether AI can make plausible music to which platforms can make that capability easiest to access, govern and integrate. By putting Lyria 3.5 into Gemini and the Gemini API at the same time, Google is betting that distribution will matter as much as raw audio quality.