Lyria 3: Google DeepMind's most advanced AI music generation

Lyria 3 is the latest generation of Google DeepMind's music AI, built to understand musicality at a fundamental level - rhythm, arrangement, instrumentation, and vocal performance. The standard Lyria 3 generates polished 30-second tracks with lyrics, vocals, and full instrumentation. Lyria 3 Pro extends this to approximately 3-minute compositions with professional-grade structural awareness - verses, choruses, bridges, and outros that flow naturally. Both versions support multi-language vocals, genre-spanning creation, and detailed prompt control over every musical element. The model was developed with input from producers and musicians including Wyclef Jean and Yung Spielburg.

Lyria 3 generates music from text prompts — describe anything from "an upbeat birthday tune" to detailed specifications with tempo, key, and instrumentation preferences. Image-to-music converts uploaded images into custom soundtracks that match the visual mood. Define realistic vocal styles and acoustic preferences to shape your sound. Lyria 3 Pro adds full structural control: outline song progression in your prompt, control when lyrics start and end, and produce tracks with verses, choruses, and natural transitions up to 3 minutes. Every output receives SynthID watermarking for AI-origin detection.

Lyria 3 is now live on Picsart's AI Playground and will be available in Flow soon. Creators can generate AI music directly within Picsart and pair it with videos from AI Video Generator, creating complete audio-visual content without leaving the platform. This connects Lyria 3's music generation with Picsart's image and video tools for a unified creative workflow.

Lyria 3 stands out as the most musically aware AI model available. It understands song structure, not just sound generation. While other models produce short clips or loops, Lyria 3 Pro generates full 3-minute compositions with natural musical progression. Multi-language vocal support, genre versatility from classical to hip-hop, and prompt-level control over arrangement give creators professional-grade output from a text description. SynthID watermarking ensures responsible use, and the model's training on authorized materials means creators can use outputs confidently. Inside Picsart, it completes the creative loop: generate visuals, generate audio, publish - all in one place.