Sound | Figuration is a live interactive piece for piano and live multimedia. In this composition, the composer wants to grasp the projection of four-dimensional objects: sound, in the three-dimensional world through a combination of various media. The sound part of piano material is generated by Magenta (Tensorflow) using a pre-trained model and reconstructed by the composer. All electronics and video parts are real-time transformed and generated by the live pianist’s performance. Therefore, the pianist’s performance and live processing audio and video create lively intermedia experience to reveal the figuration of sound in the process of emergence, growth, and distillation. This is also a brand new attempt in mixed electroacoustic music creation using machine learning.