Assuming you can make sharp and accurate clips and move the layers in a plane and up/down in the layer stack, I give the following list:
BEFORE: Be sure that
every already clipped element (=PNGs or PSDs for transparency) have a good and sharp enough edge. Fuzzy or wrongly clipped edges can be a pain when you're trying to embed elements into the composition.
all images have equal or nearly same pixel densities, for example, 300 DPI (=dots per inch). Otherwise, the new element can surprisingly be much too big or too small when clipped out and pasted into the composition.
A hint: Do not resample when you change the pixel dimensions. You will lose information. If some image, when converted to the right DPI, has much too small dimensions in millimetres or inches, it can be useless. An element can well be made smaller, but enlargening it 50% or more often renders it fuzzy. This is a mathematical fact for bitmap images.
There are some special enlargening tools such as One1 Perfect Resize, but they are costly and they can only create an illusion of sharpness, they can't find nonexistent details.
EDGES: A new object must not be sharper or fuzzier than the environment seemingly at the same distance. This is true also for the clipped edge. Often the edge or whole area of a distant or in the "out of the focus" -area standing element is too sharp and needs some blurring. Element's wrong luminosity or different colour balance increases the unreality of the edge
DIMENSIONS, PLACEMENT AND ORIENTATION: This really should be done first as soon as the wanted element is clipped properly and pasted into its place in the layer stack. You can move the layer as whole to put the new element in proper place. If you select an element and open Edit/Transform you find tools to
- scale the element larger or smaller
- move or rotate the element
- warp and distort element's shape
In Photoshop it's possible to select an element and define it to be a smart object. They are useful because successive geometrical transforms do not pile irreversible quality degradation. For example shrinking and enlargening back is possible with no quality loss. Unfortunately colour and other optical properties of smart objects are not adjustable.
SHADOWS: This should be done last or at the same time, when lights are adjusted (=shadow is low light)
You can add a layer, choose it to be in mode "Hard light" and paint the wanted shadows onto it as grey or black. Experiment with brush size, softness and opacity. It means a lot for the speed of getting the wanted result.
Note: Lighter than 50% grey paint increases the light and painting some colour makes everything underlying towards that colour.This is likely needed, for example when somebody stands near a big and bright coloured wall or in shadow under a green tree. You can really paint or spray lightness, darkness and colour as a paint when the layer is in Hard light mode
Another way to make shadows is to make blackened uncolored copies of the elements, flip them upside down, make them tilted, stretched, warped, and blurred. When they are made partially transparent (=opacity less than 100%) they can be realistic partial shadows.
Both shadow methods are needed!
REALISTIC LIGHT: It means correct luminosity, correct colour and correct contrast as a minimum.
A copy of an element can at first be explored as a different image which is moved onscreen near to its final position. It's is easy to add some adjustment layers for brightness & contrast, colour balance, hue & saturation and luminosity curve to find the right appearance. Adjustment layers are a delight because you don't destroy anything by using them. Everything stays adjustable. The depth of an effect can be reduced by making the adjustment layer more transparent or by adjusting layer's controls. Latter is preferable because there is one control less to be noted. After finding the adjustments you can make the same adjustments for the element in the composition.
Of course, you can use adjustment layers also in the final composition stack, but you must include layer masks to restrict the affected area not to be global. This way you retain the possibility to create radically various versions later more easily, but beware: It's easy to have finally quite many, even over 100 layers. Layer grouping and some ordering rules keep them still manageable.
There is a function called "Match Color" for forcing either the overall colour or the colour of some selected area to be same as in another element.
Under the title, "SHADOWS" stays also useful data for realistic light. I wrote you can even spray lightness darkness or colour. That is not enough for realistic light. Also, contrast needs to be adjusted.Illuminated areas generally have stronger contrasts than areas in shadow. If you want to paint or spray contrast, you should make an adjustment layer preferably for curves. First, twist the curve to cause the maximum needed contrast. Add a layer mask, fill it total black. You can spray or paint white into the layer mask where "the curve" is assumed to affect more strongly.
Hopefully, you are not too frightened to start. And after this every kind of glosses, reflections and transparencies are waiting for an opportunity to eat your sleeping time.
PS. This has been written very fast. Maybe it's improved after few days or at least the most bizarre typos are corrected.