Google Adds Lyria 3 AI Music Generation to the Gemini App with Text, Photo, and Video Prompts
By HiFi Editorial20 February 2026
Key Highlights
Google announced beta rollout in Japan of Gemini’s music generation feature using DeepMind’s Lyria 3 model, creating high-quality songs within seconds.
Lyria 3 improves audio generation with automatic lyric creation, finer creative controls over style, vocals, tempo, and enhanced realism and musical complexity.
Gemini supports two creation modes: text-to-music for genre, mood, or memories, and photo/video-to-music using visual context to compose songs.
From a signal-chain perspective, Lyria 3 reads less like a novelty generator and more like a tightly constrained composition engine optimized for short-form coherence. The notable shift is not just lyrical synthesis layered atop audio, but the tighter coupling between semantic intent and timbral outcome. Compared with earlier generative music systems that tended to smear transients or collapse harmonic detail under dense arrangements, Google frames Lyria 3 as emphasizing internal musical structure—phrasing, dynamic contour, and vocal presence—suggesting a model trained with stronger priors on song form rather than raw waveform plausibility. This aligns with DeepMind’s broader trajectory: fewer uncanny textures, more predictable midrange balance, and less of the phasey artifacts that audiophiles typically associate with early neural audio.
