This may not be the right section for this, after all I’m just thinking aloud.
Would I be right in assuming that to get an empties 3D coordinates, and values (Location, rotation, and scale) to a node would require a) C programming skills, and b) a new node type?
Basically I think having the ability to load up an image, or sequence of images in the node editor, and have a node which creates a linked empty in the 3D view-port which displays the image(s), but also takes the coordinates, and values of that empty and brings them back into the node editor would be extremely useful for compositing.
In the uploaded image the node editor has an image sequence which shows an explosion, to place this into a composite with 3D objects and/ or other 2D sequences would require the input of values into a translate, rotate, and scale node. It’s certainly possible but not always idea for lots of images, especially if these images are animated within a scene.
Now the 3D view shows an empty with the image sequence, in my idea this empty would send all of it’s property values (loc, rot, scale) back to this ‘3D translation’ node, which would alter the image so it looked the same in the final composite.
Imagine having a scene were you have lots of explosions going off and you want to have explosions at various distances and positions on the screen, moving them all over with the translation nodes would be quite time consuming, yet simply placing them around in the 3D view, and at various distances would give you the correct translations and the best thing is this would show in the composition in real-time, instead of having to be rendered, and re-rendered for alterations etcetera.
Another feature could be the automatic occlusion/ matting of the image, so if an explosion is placed amongst a lot of occluding geometry the image would be occluded wherever and object was in front of it, giving the composite and automatic mask.
Anyway just thinking aloud about something which I think could help when compositing scenes.