Not Yet: Random Mode

[alan2here]

• An earlier version of the program compared a (randomly sized!) and placed mix of (orig/style or result/style) patch pairs of various sizes until one was good enough and then faintly superimposed the patch onto an initially either (blank) or (upscaled input image) result image to gradually build up the full result.

This wasn't as good as the current approach, but it did have some fascinating properties, like turning all the blue in sky into clouds, and making trees into balls of tree texture, and probably giving everyone interchangeable faces too. It'd be good for this mode to be an option eventually, it's much more variable on compute too.