Ok, here we go…
The concepts here are not too complex, but I will explain it as detailed as possible so you can understand well what is going on.
What you should know is this: Whenever an action that has history is created, it copies the previous shape node and hides it, so that it serves as an input to the action node.
In this example, a poly sphere gets some vertices merged, and this is the result:
Notice the greyed out node called “polySurfaceShape1”. This is a copy of what the sphere looked like before merging some vertices. This untouched version is used as an input to the Merge input node “polyMergeVert1”, which then outputs its result to the shape node in use “pSphereShape1”. There is also a backward connection going on, but that is only for controlling the merging, has nothing to do with out discussion.
Now, to this merged sphere, I applied a nonlinear bend (this will simulate the rigging you may have done, I’m using only a bend to simplify the explanation).
Several more nodes are created. Ignore the ones to the right, they are only display and shading group nodes. In the center you will see the actual shape node “pSphereShape1”.
On the top left, you will see the same two nodes that we had before, the original untouched shape node, and the merge input node.
On the bottom left, you will see four nodes, which are created when the bend deformer is applied. “bend1” is the actual bend node, and the other three are auxiliares for it.
The real magic will happen with what you see between the merge node and the bend nodes. An intermediate shape node called “pSphereShape1Orig”. This is created to function as the input to the bend nodes.
Let’s take a look at how the important connections flow: Starting from the top left - “polySurfaceShape1” (the untouched sphere shape) is inputted into the merge node “polyMergeVert1” where the merging action is performed. The merged object is outputted to the magical “pSphereShape1Orig” shape node. So this shape node now has the form of a sphere with some vertices merged. This shape node - by some mystical form - is connected to the bend nodes. So what the “bend1” action node sees is a sphere with vertices merged, and it knows that it has to bend it a certain amount. After doing so, it outputs its result to the actual shape node “pSphereShape1”.
Many other connections are going on here as well, but these are the main ones we need to understand.
Now, to the point. How do we get rid of the merge input nodes WHILE keeping the merged vertices. If you remove the “merge node” as was suggested before, there will be nothing inputted to “pSphereShape1Orig”, and it will behave as a normal sphere (with all its vertices intact). BUT… if you remove the connection from the merge node to “pSphereShape1Orig”, then this “pSphereShape1Orig” will retain its “merged vertices” info, and pass that on to the bend nodes.
The end result: A sphere that has some vertices merged, has a bend deformer applied to it, BUT only has the input nodes for the bend… The merging has now been baked into the sphere - so to speak.
You can now safely remove “polyMergeVert1” and “polySurfaceShape1”, as they are not needed anymore.
In your case, what you will need to do, is to layout all the nodes and connections in an ordered manner, so that you can see exactly how information is flowing from one node to the next, and from this find out which connection(s) need to be removed. (You probably will have 10 times as many nodes as the example I posted, but same principles apply)
I truly hope this helps, let me know if you need help in any of this, I’d be glad to help
[EDIT: Fixed the links to the images above.]