Hi there,
This is indeed one of the things I've been having my own problems with, because I also get lots of minor adjustment images and even new images based on the original (like an arm with some fur when the previous one didn't have).
You may have found some post of mine about it in the forum and yes, Nate explained it happens because the pivot or "center" point of a meshed image changes when you modify their general "dimensions". The moment you delete the outer vertices, you're changing its "center".
I don't know if they'll change this behavior in some future version, but I've been using some workarounds these years. Let's explain...
What you want is that center doesn't change from one version to another. And for that, Spine needs to think both images are either identical (or symmetrically proportional, so the re-scale feature actually works well).
At this point you don't have neither of those, so what you can do is choosing one of the following...:
a) Fit the new image into the old size.
For this to work out you need to leave a generous padding space in all images. I'm not talking about 8 pixels on each side, but something like 64px.
That will will you an extra space for when you receive the new exported PNGs from the artist, you could use GIMP like this:
1.- Load the old image first, so the canvas size is set to the original
2.- Drag the new image into GIMP
3.- Move it so it matches the old image
4.- Hide the old layer
5.- Export the whole thing as PNG, overwriting the old image
Spine won't ask you anything, because the size wasn't changed, and you'll be able to modify the mesh to include the new area.
Cons of this method are that it's a manual process for every image, and you may run out of space if the change is too big.
b) Export at canvas size.
This is what I'm currently doing myself, much more flexible but comes at a cost.
First of all, you need to settle to a canvas size for the whole model with your artist. Mine was using ludicrously large dimensions on both vertical and horizontal axis. What you want is something you know it will be enough for additional future images like a backpack, a tall hat, etc., without needing to expand the canvas.
For example, models don't need much padding on the bottom because that's where their feet and typical root bone are. They may need more on the top, for some large hat. The sides are tricky and will depend on what you want. A backpack won't occupy much space, but a long tail could.
When you settle to that canvas size, you need to make a list of images with "some probability to be changed", In my case it's almost every body part, like the legs, torso, arms, feet, even the hands, etc. Also any part that derives from them, like tight-fitting clothes that go on top.
All of those you export them at canvas size. And from now on, your problems are gone because they'll always be the same size and have the same perceived center.
Of course this means you'll end up with HUGE images in which 90% of the space will be transparent. It's a performance hit for Spine, a lot more VRAM is used so you need a nice graphics card, and whatever you do you'll need to wait a bit for images loading every time you change skins.
However, from that point on your artist will be able to modify those parts without any worries if they're one pixel larger. You just overwrite for the fixes, and make linked meshes for the new parts. As easy as that, no manual tweaking needed.
As a bonus, you could still expand the canvas size in case it's not enough. BUT you'd need to expand it by a fixed percentage on all sides, so the pivot point or center Spine detects is still in the same place. Then Spine will detect a new size for the meshed images and offer you to re-scale the meshes. You can say yes because this time the new image is symmetrically proportional and everything will fit correctly. I actually didn't need to do this yet, but it's nice to know you can in case you underestimated the canvas size.
Ah, if you're worried about stupidly huge atlases, fear not. When Spine packs the atlases it won't include those huge transparent areas, so the only thing you're paying for this is a performance hit only when loading images.
Also, Spine is very stable right now working with these big images. I uncovered some memory leak issue because of this way of working of mine, and the devs quickly fixed them.
I know it's not an ideal solution, but it works for me.
