[zzyzek]

I'm working on an extension to the Modify in blocks Model Synthesis algorithm [0] and the Wave Function Collapse algorithm [1] called "Punch Out Model Synthesis":

https://github.com/zzyzek/PunchOutModelSynthesis

Here's a gallery of sample outputs from the algorithm:

https://github.com/zzyzek/PunchOutModelSynthesis/blob/main/r...

I have an online demo of the algorithm in action for different tilesets (it's a little rough, so be warned):

https://zzyzek.github.io/PunchOutModelSynthesis/

The idea is you take an example image, chop it into little segments and infer tile rules depending on the overlap. It's very much old fashioned "machine learning/artificial intelligence" (that is, without any neural networks involved). There's also a demo of tile rule inference idea here:

https://zzyzek.github.io/TileRuleHighlighter/

[0] https://github.com/mxgmn/WaveFunctionCollapse

[1] https://github.com/merrell42/model-synthesis