The algorithm creates a mosaic of overlapping random shapes. The color of each defined by some formula based on the color of the corresponding pixels in the original contained by the shape. Since the snail's outline color is thin, most overlapping shapes will only see a small part of and it will be drowned out by the other colors contained in the shape.
@madru correct me if I'm wrong.
If I haven't messed up the formula, it should be like this.

The algorithm places shapes iteratively, always choosing the one that most reduces visual error.
Early shapes (1-200) capture large features because they provide the biggest improvement.
Later shapes (200-1000+) progressively refine smaller details. Thin lines don't appear immediately because a single shape covering a 2-pixel line (representing only 4% of its area) barely reduces error. However, as large regions become well-approximated, the
remaining error concentrates along edges and fine details, causing the algorithm to place more shapes there. With enough iterations (1000-2000+ shapes), thin features gradually emerge through accumulated refinement.
The process naturally goes from
coarse → finedetail.
It's not that thin details
can't appear—they appear
later because they require many shapes to build up gradually.
More shapes = more detail! (fyi: The first version could process a maximum of 500 shapes, while the updated version can process significantly more.)
some examples:
The first one was created with “Quadratic Bézier,” the second with rotated rectangles, each with about 4,000 shapes. The 3rd activated all options and calculated about 32,000 shapes.
....to
emexes question: no pest control

