The problem with "photons" is that it's explained almost always wrong in popular-science books as being a localized massless particle, and even many introductory QM1 textbooks use this naive-photon pucture of "old quantum theory".
The only consistent description, however, is in terms of relativistic QFT, and there it turns out that photons are not localizable as massive particles are (although in the relativistic context with less accuracy than within non-relatistic QM). What's localized is the photon detector, and all you know, given the state of the em. field, is the detection probability for photons at the position of the detector. The probability distribution is given by the energy density's expectation value of the em. field in the given state.
The next point is that in many cases also the semiclassical approximation is good enough, i.e., you treat only the charged particles quantum-mechanically but keep the em. field classically. That explains the photoelectric effect as well as Compton scattering in leading-order perturbation theory accurately.
The quantization of the em. field and thus photons in the proper, modern sense becomes necessary as soon as quantum fluctuations become relevant effects. The most simple example is the first-principle explanation for spontaneous emission. Another example is the HOM effect:
https://en.wikipedia.org/wiki/Hong–Ou–Mandel_effect
Finally classical em. fields are quantum mechanically described by coherent states of high intensity.
Dimmed down laser light is not a proper single-photon source but a coherent state of low intensity, i.e., it's most probable to detect no em. field at all or a single photon, but with some probability you'll also detect two or more photons. The photon number of a coherent state is Posson distributed.
