To give you an idea how you could use this Martin.
If you want to create a sphere in clouds, you use a get position, then plug that into a "length to scaler" node, which works out the length between (0,0,0) to your get position (x,y,z). You can then use a conditional scaler to look at that length from the origin and if it's less than your sphere's radius (let's say a constant of 1000m), then the output is 1 (white), if not, it's 0. This plugged into a cloud node with the right altitude and depth values, will create a cloud sphere.
But because your natural origin is (0,0,0), then you will only see a semisphere, as the rest will be under your planet surface....
You can position your origin, by subtracting a vector to the get position, to trick the length to scaler in thinking it's measurement is away from (0,0,0). So by by subtracting a constant vector of say (0,1000,0), with a cloud altitude of 1000, and a cloud depth of say 4000 (this negates the cloud nodes internal algorithm of soft clipping the top and bottom of a cloud), then you'll have a sphere.
To play a little more with this, you could then add a perlin noise function to the get position value to trick again a modulating centre point, which will then give you a 'wobbly' spherical cloud.
To expand a little more, you could use a modulo scaler on the x,y,z to offset the origin by steps of say 2000m, and you'll get a 'grid' of wobbly cloud shapes.
Fun stuff.