Google 將 AI 音樂帶給大眾

• Lyria 3 model improvements: automatic lyric generation, enhanced control over style/vocals/tempo, and increased realism and musical complexity compared to previous Lyria iterations[1][5] • SynthID watermarking: imperceptible watermark embedded in all Gemini-generated tracks to identify AI-generated content[2][5] • SynthID Detector: capability added to Gemini to identify AI-generated music by uploading tracks and checking for watermark presence[2] • Training methodology: Lyria 3 trained on music from YouTube and Google sources with rights under terms of service and partner agreements[1] • Output constraints: Current generation limited to 30-second clips; longer formats and integration into other apps (e.g., Google Messages) suggested as future possibilities[3] • Nano Banana integration: image generation model creates custom album art alongside generated tracks[1][2]

Google and Apple's simultaneous launches signal intensifying competition in AI-powered creative tools as both companies race to capture consumer mindshare in generative AI. The democratization of music creation could disrupt traditional music production workflows and create new copyright challenges, evidenced by Google's proactive safeguards against artist mimicry. Expansion of Lyria 3 to YouTube creators globally positions Google to influence content creation standards across short-form video platforms. The integration of AI music generation into mainstream consumer apps (Gemini, Apple Music) suggests music creation will become an expected feature in productivity and entertainment ecosystems, similar to image and video generation. However, quality concerns noted in reviews—particularly regarding lyric composition—indicate the technology remains in early maturity stages. The emphasis on watermarking and detection tools reflects growing industry recognition of AI authenticity verification as a critical infrastructure need.