Honestly I can’t understand your goal!
blur to an object? You mean to the rendered image of an object don’t you? so, why “like if it were an image”?
I’m using composite nodes (maybe I’m wrong) and as I connect the image input to even just a viewer output, I get an X instead of the image.
You connected an image but what is the input? An actual image on your harddrive, or a render(F12) result? in the latter case you better use “Renderlayers” node as input instead of “Image”
Even worse, this is not a rule, because some time I instead get the image, but not the compositing result.
I get lost again here, but in any case, to output a render you need the Composite node. The viewer is well, just a viewer for the image window and the node editor backplate.
All in all, i got the impression that you probably want to use nodes for something that they’re either not intended for / capable of. Keep in mind that compositing nodes are mainly intended for the postproduction of images as you can argue from the list of inputs. They are able to modify your objects, and they tipically get in the workflow at the end of the render process.