New image formats can be confusing--do they mean bits per channel or bits per pixel? Video uses basically 3 color channels of 8 bits per pixel (2 to the 8th, 256, for 3 channels is 2 to the 24th or 24-bit, 16 millions colors bandwidth, and 32 bits with an alpha). Video is usually 8 bit, like DV formats, but higher end component/serial digital capture cards like DigitalVoodoo use higher bit depths.
Camera RAW images, perhaps used more than 16 bit or Cineon film scans by most AE users, have 12 or 14 bits per pixel recorded by the digital still cameras--which can be spread over a 16 bits in a Photoshop format or as an 8 bit JPEG with only 256 brightness levels. For more on camera raw and gamma, see the Adobe primers by Bruce Fraser. AE has long worked with Cineon files to simulate dynamic range, then added 16 bit linear color space, which has several benefits like reducing banding. 32 bit floating point preserves more detail, and recently the eLin plug-ins allow you to simulate the benefits of floating point in AE.
Much of visual effects work involves trying to make effects look “real” by capturing the real-world lighting with high dynamic range (HDR) photography techniques, often with different exposures. A great example is the web movie for "Acquiring the Reflectance Field of a Human Face" by Paul Debevec, et al. This stuff has been kicked around for years but ILM pushed Open Source code OpenEXR in 2003 and now all serious apps are supporting high dynamic range imaging. As Trish and Chris Meyer in Motion Graphics: A Different Light: "motion graphics artists--with their pursuit of surrealism--may find that breaking the rules yields a more interesting result." Yet, ready or not, Trish and Chris do show off gamma tricks and nudge us into a larger headspace. Stu Maschwitz also does this effectively in his demonstration movies to illustrate the power of HDR in the AE plug-in eLin.
For more info on gamma, see Digital Video and HDTV by Charles Poynton (see his website for free stuff); log and linear is discussed by Stu's ProLost, Mark Christiansen's Adobe After Effects 6.5 Studio Techniques, Steve Wright's Digital Compositing for Film and Video, Brinkmann's The Art and Science of Digital Compositing (Digital Compositing was once part of Shake docs). If you really want to know what float means, see What Every Computer Scientist Should Know About Floating-Point Arithmetic, by David Goldberg.
But "float" has costs: 32 bit float images require four times the memory and render times as 8-bit images, so most production is at lower bit depths. One wag on the AE-List (Mar 2005) added: "Of course, one can do this now in AE using Cineon/DPX out of a scanner or Spirit. The downside to comping in Log color space is that transparency and layer modes don't work properly, although they generally work good enough in most situations...my knock against working in linear float is that very few artists will look at the work at any other gamma, and high display gamma especially can hide quite large color errors near the black point. Working in log, using for instance the standard Cineon Linear Preview layer script, one tends to turn it on and off, and of course it renders off, so one typically sees the work in log before it goes to film. This is all my opinion, but doing DI [digital intermediates] we get to fix an awful lot of stuff that people screwed up in float because it looked fine on their 'calibrated' monitor [will Kodak Look and Display Manager propagate to the low end? -ed.]. That said, it has taken the facilities by storm in the last couple of years. We're getting looked down on like backward hayseeds with an ox-team for not ploughing with air-conditioned diesel tractor of float, where all problems are magically solved. As someone who believes in the value of creative cinematography, I also find the whole 'scene-referred' thing quite offensive, but that's just an implementation of OpenEXR, not HDR as a whole."