It's a skill, but it's not too hard to get decently good at it. It's mostly a matter of actually building your model in the real world and iteratively improving your digital model and instructions from there, and changing the camera angle for steps that need it.

Most of the time is spent fighting with Stud.io (the software) and dealing with its bugs (it runs OK in Bottles (i.e., managed Wine environments), but stupid rendering bugs mean you may have to do the final export step on a device with a different GPU).

When you also want to print a booklet of instructions, getting the generated steps neatly out of Stud.io's instruction maker and into a sane pipeline for making the print-ready PDFs takes some work (as well as CMYK conversions), but that is mostly a matter of setting up some scripts to wire up pdftk and ImageMagick and friends.