AO baking, best practices?

Fuiosg

A few pesky questions

I've been baking my AO maps in blender for awhile now, because with mirrored parts using the same UV space, I just find it easier to setup than exporting OBJs in and out to xnormal. But should I be using Xnormal, is there any real advantage?

- Is it a good idea to setup planes and other objects to artistically control the shadowing, or do you find it best to just bake the object as is?

- Is it generally better to break up a model into many parts, or only as few as needed to prevent artifacts?

Eric Chadwick

If blender makes good maps, and it's fast, stick with it. It would be good to test Xnormal though to compare render speed & output quality. You can also batch process with Xnormal, various people have made scripts to do this.

Setting up occluders depends on how the asset is going to be used. If it mostly sits on the ground, then yes. But if it's something more dynamic (a barrel that can be pushed over) then no.

Breaking up models is good for avoiding bake overlaps. Lots of people call this an exploded bake. But an exploded AO won't have occlusions where things really touch in the final model, so you can do two bakes... one exploded, and one not, then combine the two. EarthQuake has an example of this here:
http://wiki.polycount.com/wiki/Ambient_Occlusion_Map#EarthQuake.27s_Baking_Method

Prime8

I'm using Blender for baking AO as well.
Especially now that it has cage function for normal maps I don't need to include Xnormal in my pipeline anymore.
XNormal however supports anti-aliasing and I think it is faster, but that is difficult to compare since there are many options within Blender as well and they influence the speed a lot.
If the results you can achieve in Blender fit for what you want to do with the AO I would stick to it.