The WaveFunctionCollapse (WFC) algorithm, created by Maxim Gumin, is an example-driven generation algorithm that has been used initially in image generation and further in applications such as 2D and 3D video game level generation. The algorithm is powerful in creating cohesive and near-infinite output based on constraint solving, and because of its use of tile-like units, it may have significant applications in music and lyric generation. Previously, WFC has been applied in:
- Alice in Wonderland-inspired sonnet generation in Martin O’Leary’s Oisín project.
- the Piano Teacher game created by Jared Pettitt, Celeste C. Jewett, and Tamara Duplantis to generate example pieces for learning pianists.
In our research, we applied WFC to both MIDI beat and lyric generation,
specifically generating contemporary hip-hop music based on input from
Donald Glover’s, professionally known as Childish Gambino’s, studio
album Because the Internet. We created two different WFC generation
models using Python and Ableton Live as a MIDI editor: the beat
generation in rhythm.py is an original creation, using input MIDI
files sequenced from Childish Gambino’s music; the lyric generation is a
modified version of Oisín, using metric patterns from the input corpus.
We created successful output in each generation model with similar characteristics to the input, that can further be improved and examined by combining the lyrics and beats together into complete songs. This research demonstrates the potential of WaveFunctionCollapse in driving forward the field of genre-specific, example-based music generation.