I just made a 994 vertex, 1024 poly sphere that works well so far. There's no IBL and no shadow casting in this pic now but also I found now reasonable way to produce artifacts, IBL or not... Not sure how to upload more than one pic per post... How you did the following input? * * The insert picture button won't do it for me...

The weather would be a bit more work. A Brownian motion for clouds could be a good start because this fractal is pretty predictable. Turbulence with some tan() and sin() applied to it could look more realistic than fBm.

The next problem would be the pretty low limit for displacement. A cloud would have several 100m up to ~1km "displacement" in real world. Doing miniature clouds would still require a displacement of a couple 10m. If this is possible, - I have not tried it yet -, then it would slow down rendering coz Poser doesn't like a high displacement value much.
FF seems to be optimized for skin and such things with values for displacement way below 1.
Well, the first part is to have the cloud at a predictable height and location. Using U/V on the volume of a sphere is something that might be not possible with Poser. It's not even with most other apps, IIRC.
Next dirty hack would be to make a hollow sphere with many 2 sided planes towards y=0 and many single sided planes towards y=min/max because there is no 2nd side needed.
One would see only one side of effects at greater height and if looking down from above one would see nothing if there r 2 sides and they r used.
For medium heights and near ground level there can be stripes of volumes as well.
These volumes would act as physical volumes then. E.g. for fog or clouds.
This is such a dirty 1990's hack that I have to wash my hands now just for typing it...

A more elegant solution would be a script that creates volumes inside the sky dome sphere at locations determined by gray-values of the math nodes. For example the lighter gray parts of the turbulence stand for clouds and where it fades to darker gray the cloud grows thinner towards object y=0. This volume is then made partially transparent and some shader does the texturing job. This way any light from sun or from surface can easily be taken into account for reflections on clouds. Shaders without y would require the transformation of the theoretical y into the (x,y) matrix with modified riemann mapping theorem. (like it's done when rendering the final 2D pic)
But there's maybe some need for a biholomorphic map for y. Maybe isogonal mapping is enough since the orientation shouldn't change anywas.

That's now BB's job for next week. Period. :tt2: