Google releases music model Magenta RealTime 2, with Mac local latency below 200 milliseconds According to Beating monitoring, Google DeepMind has released the open-source real-time music generation model Magenta RealTime 2 (MRT2). The new model is optimized for Apple chip (M-series) MacBook and supports local ultra-low latency operation. Users can play and control the model as a musical instrument in real-time with a delay of less than 200 milliseconds through MIDI signals, text prompts, or audio clips. Musicians can guide music generation through various interactive methods. In MIDI guided mode, the accompaniment system will automatically generate orchestral and other combinations of accompaniment based on the chords played by the performer. In text to synthesizer mode, users can generate playable MIDI instruments by entering style descriptions such as "disco funk". In audio cloning mode, users only need to drag in a short audio sample, and the system can simulate the corresponding tone. The sound generation process also supports dynamic modulation through low-frequency oscillators, MIDI controllers, and even camera inputs, providing rich interactive dimensions for real-time sound design. In order to meet the requirements of different hardware configurations, the development team has launched two versions with different parameter specifications. The 230 million parameter mrt2_stall version can achieve smooth real-time inference on all Apple chip computers, including MacBook Air, while the 2.4 billion parameter mrt2_mase version has higher sound quality but requires advanced M-series Pro or Max chips to ensure real-time audio stream synthesis. At present, the inference library magenta rt and the C++based inference engine magentart:: core are both open source on GitHub, and the official also provides independent applications and digital audio workstation (DAW) plugins that can run directly on macOS. [Original link]