There are two steps in making a HDR picture, namely the capture and the processing.
Actually this is the same as with the making of all photographs, but there are differences for the creation of HDR pictures.
I will talk about how I capture images for HDR in Part I, and talk about the processing in Part II.
So firstly, in the capture, the question is, what scene or subject makes for a good HDR picture?
To me this will be one in which the scene's dynamic range is wide, exceeding maybe 5 EV, and where lots of details in the highlights and shadows are necessary for the picture you envisaged, or that it may
make a different kind of picture altogether, something not ordinarily seen or photographed.
Secondly the questions are what exposures to capture and at what EV interval?
From experience, I determine this by metering the significant highlights and shadows - by using the camera's spot metering - and step through this measured EV range at 1 EV interval.
For example if shadows at, say f8, are metered at 1/30 and highlights at 1/1250, then I will shoot at 1/30, 1/60, 1/125, 1/500 and 1/1000. Usually I avoid underexposure for reasons to be explained in Part II.
Further I will shoot on a tripod and use the auto exposure bracketing feature that is available on most DSLR, ie I will center at 1/125 and set a +/- 2EV bracket around it, and capture the scene at 5 different exposures in rapid mode. If you have a remote control, you may want to use it too.
The reason for these are to minimise movements/changes in the scene in the the time interval between shots, and to minimise camera movements between the shots too.
You may then ask why 1EV interval and not say 0.7 or 2 EV. Then I say, from experience, 1EV, or more, seems to work best for Photomatix when processing the HDR. You may like to experiment and verify this for yourself.
And that is all for the capture, except to add it is also possible to do HDR pictures in which objects within it moves between shots, eg waves, ripples, clouds, people, etc. There is no difference in capturing the scene. The processing software will detect and make corrections for some of these movements. However, it does not work sometimes, and then you will get artefacts.