Best example I can probably give you is with grass:
A grass mesh is made up of photorealistic meshes generated from an atlas. You 4Up the meshes so as to focus in only on the internal 50%.
This makes it so the items will “tile” when photographed.
You further force tiling by adding additional parts:
Eventually - mostly through experience - you can end up producing cinematic quality textures from this simple method - and since you bake things down, it would become rather easy to produce new texture sets for different purposes by just re-arranging the items within it in an intelligent way.
To make it more clear: the grass is made up of individual grass strands which are properly textured from a grass texture atlas which was made by picking and scanning individual blades:
