Fractal jitter has a trap worth knowing if you use this for feature placement rather than just partitioning. If you domain-warp the field, the warp has to scale with the cell size. A fixed-wavelength warp just translates the small cells instead of deforming them, so the fine scales still read as clones of each other even though the coarse ones look fine.

Two others that bought us a lot: give the features a power-law size distribution, and jitter across the whole cell rather than a safe fraction of it. Undersized jitter leaves a ghost of the original lattice that you can see long before you can name it.