This is a copy of the synthesis module of our Bloompipe web application. The webapp can currently be found at bloompipe.de!

The module reads a .pkl file into the StyleGAN network and feeds an audio stream into a vector generation algorithm. This algorithm creates vectors for each frame of the final video. Those vectors form a path (through the latent space) that represents the movement of the audio material. Those vectors are then fed into the StyleGAN which finally creates the frames.

Each frame is then stored on our server and all frames are compiled into a complete video with the underlying audio.