If it's art, to induce a feeling in the audience, then I don't see why you need to restrict it to actual photos of actual things. Once you start tweaking the picture, you can make it feel like all sorts of things that it's not, or appear to be something it's not, so why not go the whole hog and just create an image with whatever tools you can? It seems like photographs some with some sense of legitimacy as being "real" even though photographers can distort how things look to convey some feeling. Susan Sontag's essay described taking many photos of a subject until they showed the right emotion. So you can make anyone look like any emotion by cherry-picking from a huge set of shots.
> I don't see why you need to restrict it to actual photos of actual things.
We don't need to, although in that case, we might think of what we are doing as digital painting rather than photography.
> some sense of legitimacy as being "real"
A photograph is a purely-2d purely-visual representation of what we inescapably experience as 3d and multi-sensory. It can be "a real photograph" but not "real".
If what we are interested in is a documentary representation then we are making some additional claims about how the "real photograph" was made.
> any emotion by cherry-picking from a huge set of shots
Once upon a time, in the age of film photography, photo-journalists did take a huge number of exposures and have someone else process the films, and then select particular frames from contact sheets. Digital reduces that cost.
However, when someone looks at a photograph, they bring all of themselves and a little of the photograph.