Popular 720nm filters such as the Hoya R72 let thru some color, resulting in "typical" infrared look where grass is yellow and sky is blue (after channel swap - or the colors will be reversed turning sky yellow). 850mn filter however yields thoroughly black and white image, since the color filter array atop of the sensor is wholly transparent at these wavelengths - turning foliage white and skies and water black or dark shade of grey. Incidentally this means that a 850nm filter can be used to set a UniWB white balance for the camera; it's a good idea to save that WB setting in of the camera's banks for later retrieval.
This post mostly discusses using 850nm filter, since using and post-processing B&W infrared shots is easier. Filters can be acquired for around 20 euros like this 850nm one from Fotga. Results with 900nm - 1000nm filters should look similar, but the exposure times might get longer since the filter on sensor will block more and more of the light.
(Nikon AF-S 50mm f1.8G wide open.)
Lenses for infraredNot all lenses are created equal for infrared photography. Older and simpler designs tend to perform better at infrared wavelengths, less glass elements there are, the less there reflections off them. Personally I've tested some lenses:
* Nikon 18-55mm f3.5-5.6 II (the all plastic kit lens): good lens for infrared if mechanically the build quality leaves lot to be desired. No hot spot but flares quite a bit when pointed in the direction of the Sun.
* Tokina 11-16mm f2.8: unusable for infrared if fine lens otherwise, strong hotspot in the middle. Flat field correction might fend off some of the hot spot, but that's untested as of now.
* Samyang 35mm f1.4 and its variants: mixed experience, sometimes there's hotspot, and sometimes not. Flares easily. Worth a try though, it's optically marvellous lens.
* Nikon AF-S 50mm f1.8G: no hot spots, quite a bit of flare. Panoramas on this post are all shot with this lens.
* Nikon 105mm f2.8 VR Micro-Nikkor: no hot spots, "slow" aperture can make focusing difficult on newer bodies.
Kolarivision.com maintains an extensive list of lenses and their hot spot status. Surprisingly, they claim that the 50mm Nikkor and the Tokina ultrawide zoom in the above list is a bad lens for IR photography. Maybe the results depend on camera too, since the coatings and filters above the sensor can change between makes and models.
(Exposure times were 30 seconds for each image, this caused the passers-by to disappear from the image.)
Shooting and post-processingOne thing that takes patience when doing infrared photography is focusing. Since the filter only passes thru light that's invisible to the naked eye, live view must be used to focus. And the sensor's being not very sensitive to infrared, so using bigger apertures help by passing more light. For example, by f/4 even the D800's live view mode becomes useless for focusing because the signal gets drown under the noise. And that's at noon, when the Sun is brightest - even bigger apertures are needed when overcast or it's otherwise dimmer.
Some - mostly older, manual focus lenses - have an infrared focusing mark in the distance scale of the lens. It's usually small red dot or line next to the visible light focusing mark. In a pinch it can be used to perform scale focusing. None of the lenses mentioned in the list above such markings, sadly.
Next thing causing grey hairs to the photographer is the white balance, which if left on auto, will be way off. Easiest way to fix this is to take a white balance reading off from grass or tree, since they should appear as white. For some reason, Adobe's Camera raw can't adjust the white balance so much that it could be corrected in post. However, for example Raw Therapee has no such limits in its white balance tool. Since the images straight off the camera are quite dull, adding more contrast is usually needed.
(Setting the white and black point correctly adds some nice contrast to the otherwise bland image.)
Theoretically a camera equipped with Bayer filter souldn't need demosaicing if the filter is equally transparent at each pixel and color. However, no filter is perfect meaning that some shorter wavelengths are passed thru the filter creating faint false colors in the image. These faint colors make passing the pixels without demosaicing impossible, creating bad artefacts in the final image. This I tested with Raw Therapee by setting the demosaic method to none (Method=none under heading RAW) in the .pp3 file.