Google DeepMind has officially moved its latest music generation technology into the hands of developers with the public preview of Lyria 3. This release includes two specialized variants: Lyria 3 Pro, capable of composing full-length tracks up to three minutes, and Lyria 3 Clip, designed for rapid 30-second loops and social media assets. These models represent a significant leap in musical coherence, ensuring that a song's structure remains harmoniously consistent from the opening verse to the final chorus.

Beyond text-to-audio prompts, Lyria 3 introduces sophisticated controls for granular influence over the output. Developers can now use "tempo conditioning" to set specific beats per minute or "time-aligned lyrics" to dictate exactly when vocals should enter a track. Most impressively, the model supports multimodal inputs (the ability to process different types of data like images and text), allowing an image to serve as a stylistic reference for the mood and atmosphere of the resulting audio. This bridges the gap between visual inspiration and auditory execution.

Safety remains central through the integration of SynthID. This digital watermarking technology embeds an imperceptible layer into the audio, allowing anyone to verify if a track was generated by AI even after the file has been edited or compressed. By balancing creative freedom with robust attribution, Google aims to position Lyria 3 as an additive force for human artists rather than a replacement. The models are currently accessible via the Gemini API and a dedicated playground in Google AI Studio.

Google DeepMind has officially moved its latest music generation technology into the hands of developers with the public preview of Lyria 3. This release includes two specialized variants: Lyria 3 Pro, capable of composing full-length tracks up to three minutes, and Lyria 3 Clip, designed for rapid 30-second loops and social media assets. These models represent a significant leap in musical coherence, ensuring that a song's structure remains harmoniously consistent from the opening verse to the final chorus.

Beyond text-to-audio prompts, Lyria 3 introduces sophisticated controls for granular influence over the output. Developers can now use "tempo conditioning" to set specific beats per minute or "time-aligned lyrics" to dictate exactly when vocals should enter a track. Most impressively, the model supports multimodal inputs (the ability to process different types of data like images and text), allowing an image to serve as a stylistic reference for the mood and atmosphere of the resulting audio. This bridges the gap between visual inspiration and auditory execution.

Safety remains central through the integration of SynthID. This digital watermarking technology embeds an imperceptible layer into the audio, allowing anyone to verify if a track was generated by AI even after the file has been edited or compressed. By balancing creative freedom with robust attribution, Google aims to position Lyria 3 as an additive force for human artists rather than a replacement. The models are currently accessible via the Gemini API and a dedicated playground in Google AI Studio.