Google has put DeepMind's Lyria 3 into Gemini, letting anyone 18 or over make 30-second songs with vocals, lyrics and cover art from a text prompt or an uploaded photo or video. It launched on desktop first, in eight languages.
Google says artist names in prompts are treated as loose inspiration, that outputs are checked against existing content, and that every track carries a SynthID watermark. It has not disclosed the training data or deals behind the model.
The Decoder places the launch amid label lawsuits over AI music and notes reports that OpenAI is building a rival generator.
Sources · 1