Back on the topic of ND grads...
The grads typically come in 1 stop, 2 stop, and 3 stop versions... and they also use have an option for a "hard edge" vs. a "soft edge" grad.
The 1, 2, and 3 stop difference indicates how much light is blocked by the half of the filter that is tinted. The idea is that you'd meter the light area (presumably that would be the sky) then meter the dark area (presumably that would be the foreground) and note the difference in the exposure. This tells you which grad strength to select.
Next is the issue of "soft" vs. "hard". Both filters have a gradual transition (hence the name "grad") but a "hard" grad has a rapid transition from clear to dark whereas the "soft" has a much slower transition. If you're shooting a scene where with a very clear and clean separation of foreground vs. sky (the scene has a horizon line) then a "hard" edge grad usually is the better choice. If the scene has no strong "horizon line" (elements from the foreground cross the horizon and reach into the "sky" part of the shot) then a "hard" edge grad would look wonky because you'd notice foreground objects having a very obvious transition point. A "soft" edge grad is gentle enough that the viewer probably wont notice the objects are darkened near the top (or whichever direction you decide to orient the filter).
For a full-frame camera, the 100mm filter width (4") is the appropriate size (they do come in sizes).
You'll need a filter holder... Cokin makes them and Lee Filters make them. Lee Filter is a high-end brand, their filters are better and their holders have more options. Cokin will be cheaper (they're a budget name) and they wont have as many options. However... for the basic 100mm (4" width) holder, they'll both work well.
Lastly... can you do this in post-processing or do you need the real filter?
That last question is really an issue of total dynamic range. What you definitely don't want to do is allow the information in your image to "clip" (aka "blown" highlights is usually the problem.) That means if you look at the histogram that represents the distribution of data in your image, the histogram often resembles something that looks like the silhouette of a mountain range on a graph. If one side of that "mountain" is chopped off (either at the left or right edge of the histogram) then it means you lost data that can never be recovered (that's something you cannot fix in Photoshop... unless you're a talaneted artist and can make a really convincing fake.)
This means that if your camera has enough dynamic range to capture whatever the scene needs, then you don't need a physical filter. If the exposure range of the scene exceeds the dynamic range of the camera then a filter would be helpful as it buys you several additional stops worth of range.
Ernie mentioned the use of HDR (HDR = High Dynamic Range photography). In HDR you should use a tripod and the subject should preferably not be moving. You'll take several photographs of the same subject... but each photograph is taken using a different exposure. One of the photographs should be taken using the overall recommended exposure (if you use "evaluative metering" then you use the suggested exposure). Typically you would take at least two additional photos... one is usually deliberately over-exposed by a few stops (often 2 or 3) and the other is typically underexposed by a few stops.
You can then "merge" the three shots. The camera evaluates the middle exposure and finds the brightest and darkest pixels. It then borrows pixels from the over-exposed and under-exposed versions of the shot to integrate into the final exposure and this creates a result where nothing is over-exposed or under-exposed.
I'm not sure about the 6D, but several cameras have a built-in HDR mode. There are many options as to how you process an HDR image and some look rather surreal while others look natural. It is possible to make HDR exposures that look very convincing (they don't "look" like HDR shots) and this is a way to get the shot without using filters at all.
But the one caveat to HDR is that it really helps if nothing in the scene is moving. If objects are moving through the scene then you don't get three (or more) shots with identical scenes (just different exposures) but you end up with three completely different shots that creates "ghosting" problems (like a double-exposure) when you try to merge them. Some of the better software has anti-ghosting algorithms that try to detect and correct this. Since a shot taken with a grad ND filter gets everything in a single exposure, object movement in the scene is not a problem.
Tim Campbell
5D III, 5D IV, 60Da
