While the debate will certainly rage on about the quality and usability issues surrounding compressed video, there are those of us who barrel on ahead and finish projects at Avid Online Video Resolutions, artifacts and all. To that end, I've prepared a
list of suggestions for my graphics freelancers which some of you may find useful. The list is pretty specific to my needs, so please feel free to comment and/or add to the list:
JPEG compression functions by reducing the overall frame into a grid (usually made up of 8x8 pixel boxes) and then selectively limiting the color gamut for each box based on the compression ratio. Because JPEG is a "lossy" compression method,
this means that color information is thrown out and replaced by the algorhythm's best guess at an accurate match for that block. Even at higher quality settings, is not uncommon to see boxy JPEG artifacts appear in passages of a frame that contain lots of
chroma or luminance detail. This can result in a noisy halo effect around formerly crisp-edged titles, banding in otherwise clean gradients between colors, and other unpleasantness.
Here are some suggestions for avoiding problems with Avid's JPEG codec: Avid's JPEG routines seem to be primed for preserving luminance detail over chrominance. Therefore, graphics that contain wide fluctuations in grayscale information seem to hold up
better than those which contain a wide color gamut. For instance: white titles over a neutral gray or black background seem to import better than a light blue title over an orange background of the same typeface and typesize. While the design of graphics
that are fairly monochromatic or bi-chromatic is not appropriate in many instances, the do tend to hold up fairly well, and should be used whenever possible. If the overall design requires harsh color contrasts (f.e, blue over orange) try to limit the sat
uration of both colors as much as possible.
1. Background plates that contain clean gradient fills tend to experience "banding" or other artifacts, especially at those points in which the algorhythm has trouble resolving a wide range of chroma values. In general, desaturated and darker
backgrounds hold up better (f.e. dark blue to black.) Gradients from one luminance value to another also hold up fairly well (f.e. medium blue to dark blue.) Those gradients which have color to color transitions (f.e. yellow to red) don't take compression
as well. Limiting the chroma and modulating the luminance usually works better.
2. Paradoxically, adding texture to a gradient background plate (f.e. blue to black with a bit of added PhotoShop noise) will usually smooth out the compression banding. This is due to the fact that noise and artifacting added by compression will tend t
o blend-in with the textured graphic. Mono-chrome or chroma-limited textures such as granites and marbles work well as background plates for precisely this reason.
3. Backgrounds that have their origin in reality can also work well, as long as the chroma and luminance information is fairly muted. Backgrounds created from still-video frames, scanned photographs, stock photos, etc. often take compression fairly well
when treated as background elements. It seems that the complexities of textures that occur in these material often have the same effect as added noise. Again, limiting the choma and luminance variations (such as the application of PhotoShop mono or duoto
nes) yields the best results.
4. Sometimes a slight Photoshop blur over areas of complex details in a background plate will even out the color gamut of a given passage, and will allow it to import more accurately.
1. By nature, typographic text contains areas of high contrast from pixel to pixel, which allows for maximum intelligibility. Unfortunately, JPEG compression schemes perform least satisfactorily under precisely these conditions, and it can become a chal
lenge to design graphics which do not fuzz-out or grow noisy or blurred edges when imported into the Avid. Like background plates, typefaces and type attributes should be chosen which minimize pixel to pixel contrast, especially in the chroma. There is, o
f course, a continuum between clean compressibility and intelligibility. When designing for JPEG systems, your job becomes finding that happy medium.
2. If appropriate to the design, try to pick colors which do not vary widely from the information in the background plate (f.e. light blue text over a blue gradient.) Bear in mind that variances in luminance hold up better than color contrasts (f.e. use
white instead of red text over that forest-green foliage shot) White text is usually a fairly safe bet, as it contains no color information (or all of it, depending on your perspective.) While it's good to have common sense and a basic understanding of c
olor theory, simply using the Apple color picker to select colors which are fairly close together-yet still intelligible-works just as well.
3. When appropriate to the design, semi-transparent text generally imports well (f.e. 85% transparent white text over a desaturated gradient background.) Avoiding the use of 100% white text (try 90% white, 10% black) or fully saturated colors will also
help to limit the constrast between pixels, and is not too terrible a sacrifice for intelligibility.
4. Soft and transparent drop-shadows can get "swallowed" by compression-especially those which occur over fairly dark backgrounds. This is probably due to the fact that the soft edges of a soft drop-shadow get averaged into their overall surro
undings, and distorted by the codec. Experimentation and experience with varying degrees of shadow transparancy, blur, pixel spread, etc. is the only way to guarantee accurate representation. Try importing several graphics with shadows before you commit t
o one method.
5. Logic dictates that larger, simpler typefaces hold up better than smaller, more detailed fonts, and experience bears this out. Small ligiatures and florid brush-strokes can really get messed up by compression. In other words, check out that wedding i
nvitation font before you use it.
Keyed titles and other elements behave in exactly the same way as other compressed graphics, with one twist: You lose precise control over the color gamut of the background, unless you can colorize or otherwise alter whatever footage you're keying over.
As it's not likely that the CEO you're titling will necessarily want to be tinted blue for a few seconds, your titles will have to make up for it:
1. The suggestions for text described above (limiting contrast, color gamut, etc.) apply in the same manner, but you must keep in mind the background footage. This can be complicated by needing to design graphic keys to fit a wide variety of footage, or
keying over pans, zooms, etc. When in doubt, stick to desaturated color in text keys.
2. Exporting a PICT frame or two can allow you to check your color choices against what is in the immediate background. Additionally, it is also possible to select exact color values from the exported PICT image that complement the design in a pleasing
manner (f.e. picking the exact color of that mauve wall-paper above the talent's head for the title.)
3. Semi-transparent boxes or gradients behind fully opaque keyed text (f.e. a lower third name slug with a 50% transparent black background) greatly help to reduce the color-gamut contrast of the background pixels. It imports fairly cleanly, and it's hi
p and nineties, too. Make that alpha-channel work for you!