Crates.io | wfc |

lib.rs | wfc |

version | 0.10.7 |

source | src |

created_at | 2018-12-23 14:45:21.014336 |

updated_at | 2022-12-24 08:46:07.908157 |

description | Generate images using Wave Function Collapse |

homepage | https://github.com/gridbugs/wfc.git |

repository | https://github.com/gridbugs/wfc.git |

max_upload_size | |

id | 103417 |

size | 76,658 |

https://docs.rs/wfc

Library for generating grids of values which are similar to a specified grid. A typical use case for this is procedurally-generated images, though it generalizes to any grid of values.

*Similar* is defined as both:

*strictly locally similar*: every small (typically 3x3) pattern in the output image appears somewhere in the input image.*loosely globally similar*: the distribution of small patterns in the output image is roughly the same as the distribution of small patterns in the input image.

Grids are populated using a constraint solver. For each cell, we store a probability distribution representing how likely that cell is to contain the top-left corner of possible pattern. Initially the probability of each pattern is based on its frequency in the sample image. Then, it repeatedly identifies the cell whose entropy is the lowest, and decides (randomly, weighted by probability distribution) which pattern to assign to the cell. This assignment may remove some candidate patterns from neighbouring cells, so it then updates candidate cells. This process of choosing a cell, assigning it a pattern, and propagating incompatible neighbours continues until either the entire grid is populated with values, or all the candidates are removed from a cell.

->

For more image examples, see wfc-image.

This shows the process of generating an image based on the sample flowers image above. The colour of each pixel is the average of all colours which could be assigned to it, weighted by probability.

- Maxim Gumin's WaveFunctionCollapse is where I first learnt about the WFC algorithm. It contains a reference implementation and collects links to many other WFC resources and implementations (including this one).
- WFC is heavily based on Paul Merrell's Model Synthesis algorithm.
- I found Fehr Mathieu's fast-wfc to be a very understandable implementation of WFC. It answered many of my questions about specific details of the algorithm.