6:31 pmOff
Lens Movements
There is quite a bit of confusion—at least in the terminology—regarding the lens adjustments made possible by perspective control lenses (also called "tilt/shift" lenses). The best, thoroughest, clearest explanation I've come across for this part of the photography technique comes from chapter 10, "View-Camera Adjustments", in Ansel Adams' famous "The Camera" book. I recommend you look it up for a more extensive explanation complete with illustrations and examples.
What gives these lenses their potential is that they have a larger image circle than regular lenses. Instead of simply being large enough to cover the sensor, their image circle extends far beyond and can therefore be moved around without introducing vignetting (not to be confused with illumination fall-off).
Geometric Distortion
The first set of adjustments are made to control convergence.
For example, this is what happens when the top part of a building appears smaller than the bottom part, because of a viewpoint that forces the camera to be tilted back. You cannot adjust the perspective with lens movements (only physically moving to a different position can change the relative distance the camera stands from the bottom and the top part of the building), but you can do something about the converging lines.
The basic idea is to position the camera so that the sensor plane is parallel to the surface you want to keep straight and simply slide the lens in order to place the part of the image circle that contains the subject where you want it. The amount of correction is therefore limited by how large the image circle is.
The sliding of the lens can be made in any direction allowed by the lens, to correct for lines converging in various directions (convergence doesn't necessarily only happen when looking up!) You can also create high-quality panoramic images by sliding the lens between exposures (never actually panning the camera), to produce (almost) seamless stitches.
If the lens is slid upwards, the movement is called a "rise". A downwards slide is called a "fall". Sideways, it is called a "shift" (left or right).
Focus Plane
The second set of adjustments are made to control the focus plane.
With regular lenses, the sensor plane and the lens plane are parallel, resulting in a parallel focus plane (depth of field extends front to back, parallel to the sensor). By changing the angle of the lens, the Scheimpflug principle explains that the plane of focus will end up at an angle as well, with the depth of field extending like a cone around that plane (see article for a complete explanation).
You don't need to bother with all that math. In simple terms, what it means is that because you can change the angle of the plane of focus, you can exert more control over which part of the image will be sharp. You could, for example, achieve a sharp image from right in front of the lens to the infinity, without requiring an overly small aperture that would be at worse impossible with regular lenses, at best detrimental to image quality because of diffraction.
By exploiting this optical phenomenon in the "wrong" direction, you can create a very slim section of sharpness in the image (almost perpendicular to the sensor plane), with the rest falling sideways out of focus — the trick behind the "fake miniature" effect.
When the lens is angled up or down, the movement is called a "tilt". When the lens is angled left or right, the movement is called a "swing".
Because the lens can be rotated, a combination of rise/fall/shift/tilt/swing can be achieved (depending on the capabilities of the lens — recent designs allow the slide movements to be rotated independently from the tilt/swing movements).
2:25 amOff
Gel That Flash
When you want to harmonize the color of your flash with the ambient light, you first have to know which color the ambient light is, obviously. You might try to go at it by eye, or even by tradition, but you might not get the result you expected (for example, fluorescent light used to be very green, but this is not necessarily a reliable presumption anymore, since they now come in many colors, with arbitrary names like "cool white").
Since we don't walk around with colorimeters or computers to analyze our images (and because we just don't have the time to fool around anyway), there is an easy trick to reliably figure out which color the ambient light actually is. Your camera has an RGB histogram — all you need to do is to fill the frame with a neutral reference and you'll know!
Remember that since you only want to read the ambient light for this test, you need to turn off your flash. Also, it is important to use the "daylight" (neutral) white balance to do this test, otherwise you will not get a true assessment of the color (you will get something that was corrected in one way or another). The neutral reference could be a grey card (ideally), but it could also just be a white wall or piece of paper — as long as it fills the frame and is "close enough" to neutral you'll be alright, since you won't be able to surgically match your flash to that color anyway (there's only so much you can do with a couple of gels!)
I tested this in my living room, where the lamps use "compact fluorescent" bulbs — those are hard to guess. When the image appears on your LCD, display the RGB histogram, and you will see the color dominance right there:
In this case, we can see that there is far more red and green than blue. We can also see that although this is fluorescent, this is definitely not just green. Remember that with light, red+green=yellow! So in this case we have some kind of orange, because the red component is a bit stronger than the green.
Now all you have to do is pick gels that will add the same cast to your flash. If I had to act quickly, I would pick a strong orange and I know that I would get a pretty decent result. With just this one quick test, I am certain not to make a huge mistake and use a color that is way off.
To illustrate the effect the different gels produce on your flash, you can do a similar test: shoot a neutral reference, in "daylight" white balance, but this time with only the flash exposure (use a fast shutter speed and low ISO to remove any ambient light).
Using a strong green gel, we get this result:
Definitely not! Notice how the histogram doesn't look at all like the one I had with the ambient light only. Let's try a strong orange gel:
Well, this is much better. I could certainly shoot knowing that the color of the ambient light and the color of my flash are "close enough". If you want to match the ambient light even more closely, it's possible — at some point you'll get something like this:
A-ha! Well this is very, very close. But I had to use three different gels (from what I had in my kit), so this is getting a bit crazy. Also remember that each gel you add in front of your flash cuts some light — the more gels you use, the less powerful your flash becomes.
In practice, you might only have 2 or 3 degrees of orange and green in a basic kit. That's fine — most discrepancies can be removed with just these, even if they are not "perfect". The more you do this, the more you will be able to pick the appropriate gel (or combination of gels), since you will know what histogram they produce.
6:55 pmOff
Mixing Flash with Ambient
Setting the shutter speed in the light meter
When using a Sekonic L-358 light meter to read a flash exposure (the same applies to other models, I am simply not familiar with them), one of the nice features it has is its ability to indicate which percentage of the total exposure came from the flash — the rest of the exposure being from the ambient light available in the location where the metering was done.
Here is a figure taken from the L-358's user manual:
Metering flash exposure
As you can see in this example, 70% — "Percentage of flash in total exposure" — means that the flash was responsible for 70% of the total light, while 30% came from the ambient light available. The total amount of light ended up requiring an aperture of f/5.6 and 3/10th (let's call this f/6.3).
If 30% of the total light gathered by the light meter during a reading comes from the ambient light, if you change the shutter speed (without changing the flash power), you should expect that percentage to change: the longer you expose, the more the ambient light will have an impact on the total light, since you will gather more of it (while still gathering the same amount of flash). On the other hand, if you expose for a very brief moment, you will gather very little of the ambient light (while still gathering the same amount of flash), so the ambient light won't have much of an impact in the total light.
You might very well end up in a situation where the flash is only responsible for a small fraction of the total light (say, 10%) or, at the other extreme, end up in a situation where the ambient light has no influence on the exposure (which would give a 100% flash exposure). For example, when working in a studio environment, we usually don't mind leaving the modeling lights on when shooting, because we know that the light they produce is insignificant in the total exposure compared to the flash power (we get a 100% flash exposure all the time regardless).
Now, coming back to a location shoot where we mix some ambient light with flash, we should often expect situations where the percentage of the flash in the total exposure will not be 100%. In the situation of the figure above, where 70% of the total light came from the flash, suppose we were to expose for 1/30 instead of 1/125 — that is two stops more ambient, giving us a brighter background. In this case, the flash would now only account for ~40% of the total exposure. What's more, since we're adding ambient light to the previous exposure (flash remained the same), we now have more light overall, so we must use a smaller aperture — we go from f/6.3 to ~f/9.
This makes the shutter speed a crucial parameter to provide to the light meter, otherwise the obtained value will be wrong.
The only situation in which we could ignore the shutter speed would be in a pitch-black room where the only light is the flash — in this case, the ambient light (or lack thereof) would have no impact in the reading whatsoever.
Changing our perception of flash color gels
If you're mixing flash with ambient light, you know that you can (must?) correct the color of the flash so that its light color is well harmonized with the ambient light color, lest you get a quirky result.
But you can also use color gels not necessarily because you want to make the two light sources the same, but because you want to add an effect, such as warming up the subject so that it contrasts more with a colder background, for example.
In the image below, both the foreground and the background lights are of the same neutral color:
Foreground and background light have same color
Using a CTO gel (orange) on the flash aimed at the subject, you can warm up the subject, which will further isolate it from the background (which will now be of a relatively colder color than subject, even if it is the same as in the first shot):
Foreground light is warmer than background light (note: a strong color was used to make the illustration obvious)
Indeed, while the background remained the same, it can be said that the subject is warmer.
You will notice that I also (conveniently) included a gray card in the frame. If we take the picture above (the same exact file), but that we color correct using the gray card, here's what we get:
Image above, but white balance corrected with the gray card
You will notice that the subject is now perfectly neutral — exactly back to what we had in the first frame — but that the background turned blue (colder)! In this case, can it be said that the subject is warmer than in the first shot? Absolutely not — in fact, they are exactly the same, after correction, even if a "warming" gel was used.
If you're going to include a gray card (or GretagMacbeth chart) in your frame, you might want to do it while you're not using a gel on the subject, otherwise, balancing the color with a neutral reference will remove any color you might have thrown on the subject and, rather, affect the background. If you're going to include a gray card in your frame while you are using a gel, the way you should look at it is that you are in fact changing the color of the background in the opposite direction than that of the color gel (e.g.: if you're using a warming color on the subject, you're actually not changing the subject, but rendering the background colder!)
The fact that we are using a "warming" gel has no bearing on the final look of the image, because it depends on the white balance selected. A "warming" gel might therefore be more meaningfully seen as a "gel that makes the background colder"! That is why, when you work indoors under tungsten lights, you use a warming gel: not to warm up the subject (you still want it neutral), but rather to make the background colder.
8:11 amOff
On White Balance
So you want to get accurate white balance. There are many ways to get there, if you know why you are doing it and what you want to achieve.
If you're shooting raw, you're probably aware that you don't technically need to worry about it at capture time, because it's not set in stone — whatever setting you use won't definitely affect your final image; you'll easily be able to change it in your raw processor. (Of course if you're shooting JPEG, then you'll definitely want to nail it as close as possible to your objective to minimize the amount of manipulation required. But you're not shooting JPEG anyway, are you?)
So you're shooting raw. Why even bother performing your white balance at capture time, then?
- One reason might be to get a better approximate histogram, so that you can better judge your exposure. Indeed, you might think that one of the channels is clipping when, in fact, it's the color cast that was throwing off the histogram. But then again, if you're evaluating the exposure another way (say, with a light meter), then this won't matter to you.
- Another reason might be to get a better approximate image preview, so that you can better judge the colors. Of course, you can't really rely on the image preview on that little LCD for accurate color evaluation, but surely, if the color balance is completely off, it might be quite misleading. But then again, you might only use the image on the LCD to judge the composition, pose and other aspects of the image, keeping in mind that color will be dealt with afterwards, so this, too, might not matter to you.
- Yet another reason might be so that when the unprocessed images pop up on your computer screen, the client watching you work will have a better idea of what the images will look like right away. Indeed, the client might not respond well if there is an annoying color cast in the images, even if you know you'll fix it later. But if you're not shooting tethered with a client watching over your shoulder, this might not matter to you either. (Note that some raw processors, such as Capture One, can be configured to automatically apply a certain white balance to all imported images, so that the images are immediately corrected, regardless of how they were shot, which is another reason why performing in-camera white balance might be irrelevant to you.)
Only you can know how important it is for images to be properly balanced at capture time, but if you want to do it, there are many ways to go about it.
- One way is to use the Auto White Balance feature of your camera. Sometimes, in rapidly/wildly changing light situations, this setting might be more efficient at getting a decent approximate white balance than having to fiddle repeatedly with your camera's parameters. If you're not working in such a difficult situation and the color of the light is not changing — say, you're working in a studio — then this setting might, on the contrary, produce unpredictable results.
- To get consistent results, another way is to use one of the white balance presets your camera offers — things like "sunny", "cloudy", "tungsten", "fluorescent", etc. This usually won't give you an exact result, but you'll likely get reasonably close to your target for your histogram and image preview to be very usable.
- If this is not good enough, a more sophisticated way to set the white balance would be to manually enter a kelvin temperature (which, unfortunately, only affects the blue-yellow axis, not the green-magenta color), or further tweak the result with the "white balance shift" tool. (Have fun. Who wants to spend time doing that?)
- Finally, the most accurate in-camera result you can achieve would be to use the "Custom White Balance" feature.
Bear in mind that whatever approach you choose, none — not even the custom one — will give you a surgically exact result. The only way to get a better result would be to include a neutral reference in a photo under each light situation, and later use that reference to synchronize the white balance for all the images shot under this type of lighting.
For the most accurate readings, you might include a tool like a WhiBal gray card, which is good to take care of the white balance, or go one step further with an X-Rite ColorChecker Passport, which includes a good ol' Gretag MacBeth color chart as well as the software required to generate camera calibration profiles for Lightroom — serious business!
If you don't want to spend the money (or carry these around, or shoot those test shots), for fairly good results, many common objects will be good enough to get you close to your target and allow you to work without having to worry too much about color. This might be a piece of white paper, tissue, gray hair, etc. It would be pretty hard to judge if these references are absolutely neutral (unlikely), so you understand that those are not perfect solutions. (For example, stuff usually increasingly turns yellow as it gets older.)
But keep in mind that few situations require critically accurate color — you be the judge. (Sometimes, even a perfect white balance won't be enough and colors will have to be manually tweaked in Photoshop with a virtual reference anyway (such as a Pantone swatch), and yet even then, color accuracy will likely take another hit when the image gets to the printer, so this discussion is at least partly moot.)
Now, in those situations where extreme color accuracy will not be an issue, you will have to determine if accurately reproducing the scene is what's important to you. Often, perfect color balance will not yield the image that will look better to your eye — you might very well decide that warming, cooling or otherwise coloring your image will produce something that is more pleasing, whatever light was there the moment you took the picture. You might even decide to convert your images to monochrome, and maybe even to tone them. Yet another reason why you might not need to worry too much about technical accuracy, but more about achieving an image that speaks to you.
For those reasons, on a more practical/pragmatic note, unless critical color accuracy is expected, I will usually save time and use one of the camera presets to get reasonably close results in-camera (so that I can still get useful information out of my histogram and image preview), include a WhiBal in the first shot if there are no other useful neutral references around, and tweak the results in post anyway. If I know in advance that I will deliberately change the look of an image, I might use a different preset (such as using a "cloudy" setting on a sunny day to make the image warmer); this will give me a better instant feedback, even if it won't have any definitive effect on the image until I make the final decision/adjustment in post.
If you still want to perform a custom white balance to get as close as you can in-camera, note that Canon and Nikon employ dramatically different procedures to achieve this (I am not familiar with the way other brands work):
- With a Canon camera, the procedure requires a ridiculous number of steps (which explains why I usually don't bother). You first have to shoot a picture of a neutral reference, filling the center part of the frame with it. You then have to go to the "Custom WB" menu option. You then have to point the camera to the picture of the neutral reference you want to use for the calibration (usually the one you just shot), and then confirm your choice. You then have, each time, to dismiss a non-removable reminder that tells you to change your white balance setting to "Custom". You then have to go ahead and change the white balance setting to "Custom". (Still there?) If this wasn't enough, you'll also be left with an otherwise useless frame, on your memory card, that was shot just for this, which you might want to delete. (Or you'll keep it and use it to perfect the correction in post...)
- With a Nikon camera, the procedure is decisively more efficient (take a hint, Canon). Set the camera to the "Pre" (preset) white balance setting. Hold the white balance button for two seconds until "PRE" starts blinking. Shoot your neutral reference.
If the procedure fails because the camera doesn't want to take the shot, this is because you're in a "one-shot/AF-A" autofocus mode that won't allow you to shoot until focus is obtained. Contrary to what you might have been told, performing a custom white balance does not require the camera to be set to manual focus, but since the neutral reference is usually a plain card, the AF system simply won't be able to achieve focus, which is the reason it will appear not to work. Just focus on something else or set the lens to manual focus — it doesn't really matter, as long as the center part of the frame contains the neutral reference. (If you're shooting in a continuous AF mode, this won't be an issue.)
If the procedure fails with the camera telling you it couldn't perform the calibration, that's because the exposure was completely off: the camera cannot read the color information out of a clipped highlight. Make the exposure average, or use a semi-automatic mode just for that shot (such as aperture-priority).
Make sure you actually do your reading based on the actual light conditions you'll be shooting in... This is especially relevant when you're working with flash — you don't want to perform your studio calibration based on the ambient light, or your location calibration on incorrectly gelled/randomly bounced mixed-light and get results that don't make sense.
5:30 amOff
Another Bunch of Web Concepts
Following a previous post, I kept hearing a discouraging amount of misinformation regarding some web concepts, so I thought I'd do as I did before and add my two cents.
Maximum Image Dimensions and Page Size
Alright, we all agree that we want our website to load rather quickly, because we know people aren't going to wait more than a couple of seconds. We also don't want mischievous people to be able to do much with our images. We have to consider the browser size of the bulk of our visitors when determining the size of our website, because we don't want to force them to scroll to see the whole thing.
These are all laudable objectives — I couldn't really disagree with them, in principle.
But 100-120 KB per page at most? 500-600 pixels wide images at most? That, I'm sorry, is based on egregiously outdated standards. Just so you don't have to take my word for it, I visited a bunch of photographers' web sites (they all happen to have blogs I follow) and looked at how large their images were — not the home page image, I mean the average image they present in their portfolio. You'll likely recognize these guys' name — I think they know what they are doing and/or have been professionally counselled.
The smallest images I found were those on Chris Orwig's site, and they were 693 pixels wide. Here's the rest of the random sample I visited:
- Joe McNally, 778 pixels wide
- Zack Arias, 804 pixels wide
- James Rubio, 819 pixels wide
- Drew Gardner, 826 pixels wide
- Chase Jarvis, 920 pixels wide
- Tim Tadder, 940 pixels wide
- Vincent Laforet, 1024 pixels wide
- Joey L, 1064 pixels wide
- Finn O'Hara, 1278 pixels wide
Wait, what? 1278 pixels wide? That's just the image, not even the website itself.
Do you have an idea what a single portfolio section of a site presenting such good quality imagery must weigh? A lot more than 120 KB. Did I have to wait an eternity to see the images, so much so that I thought I should give up and go to another site? Not at all. Do these guys really worry about their images getting stolen? Puh-lease.
I think I rest my case.
More on Image File Formats
Okay, so I'd already covered the key points in my previous post, but let's add some more information.
The idea that the PNG format came from PCs and never really caught on on Macs is simply baseless. The main reasons the PNG format was created were to improve upon the limited GIF format and to get rid of licencing issues (since GIF was patented by CompuServe). If anything, the Linux/free-software/open-source community (which was later, even before the OS X transition, largely embraced/encouraged/supported by Apple) did more for the format than the Microsoft-riddled make-everything-proprietary PCs. Indeed, Internet Explorer only recently got the memo that PNGs had an alpha channel, which is likely the main reason the format couldn't fully be exploited in the first place, hindering it's spread.
The idea that logos, which often use few basic colors, are prime candidates for the GIF format is only partially true. When you create shapes (be it in Photoshop or Illustrator) and convert them to raster images, new intermediate colors have to be interpolated to create seemingly smooth lines and curves — a process called anti-aliasing. The consequence is that even if you're only using few base colors when designing your logo, it might very well end up requiring more colors than can fit in an indexed color file format such as GIF. To be safe, go PNG, which is just as size-efficient, but not 8-bit palette limited.
The suggestion that you shouldn't embed color profiles in images to reduce file size is to be taken cautiously. If we all agree that images on the web should all be converted to the sRGB color space anyway (because we cannot presume color management will be available in the client's browser) and therefore don't decide to include the profile in our web images, well, it shouldn't really be a problem — if the profile is missing, every browser should presume sRGB, just as if color management was unavailable, so we're good. The risk, though, is that we some day forget that an image we export is for some other use (be it for a client, for a commercial printer that color manages, etc.) and then wonder why our images don't look as expected. Seriously guys, a color profile is about 2 KB — that's insignificant in comparison to the whole size of an image. Remove one step in your workflow and always embed color profiles, just in case; it's not going to hurt to do it, but it might if you don't.
The suggestion of creating a slideshow of photos using Photoshop's Animation functionality and exporting it as an animated GIF is preposterous even as a simple example. GIF is, as we know, a monstrously awful file format for photos, because it is 8-bit palette limited and will introduce screaming posterization or dithering. Furthermore, it will create huge file sizes because it is not designed to handle photos. Additionally, it won't allow (in any remotely economical fashion) transitions between images. Moreover, it won't allow any kind of control for the user (stop, pause, forward, back). Frankly, anything else would be a better idea — Javascript, Flash ... you name it.
Output Sharpening
I've already talked about the mistaken idea that the "Save for web" functionality would export smaller files. But what preparing images through Photoshop carelessly will also not do for you is screen output sharpening. Just save yourself the time and effort and error-prone procedure of doing that by hand the olde way and use something like Lightroom. (Or at least create yourself a batch action that also performs basic sharpening.)
Spacers
The dreaded "spacer.gif" (yes, this is the common name, not "single.gif") is a bronze-age-old hack. A trademark of poor code, of lazy development. There are other (better) ways to achieve what this shameless little bugger is doing.
If you don't mind the contempt you'll get for using it and still want to go ahead with the idea, keep in mind that images can be sized any way you want on a web page. This means that you should only create a single one of those darn things, one that is 1x1 pixel, and size it according to your needs on the page. For example, if you need a 600x400 empty placeholder, don't create a separate GIF file for the placeholder, just size your spacer pixel 600x400 at that location!
Alternate Text
You know the "alt" attribute of images? The idea behind that has absolutely nothing to do with search engines. Yes, of course, search engines will use whatever additional data at their disposal to enrich their databases and, hopefully, provide more accurate results. But that was not the motivation behind the "alt" attribute (which is required in strict versions of HTML).
The name says it: it is an alternative to the image, in case the image cannot be loaded or otherwise consumed. If the image is missing or there is any kind of problem, instead of showing the image, the browser will simply display the alternate text. If a blind person visits your website, not being able to see the images, their text-to-speech software will read the alternate text aloud. Therefore, the alternate text should be descriptive of what the viewer cannot see, not just a useless generic name like "image"!
Also, if you're using a spacer pixel (goodness forbid), you should include an empty alternate text (alt=""), otherwise if the text were to be called upon, it could break your design.
General notes on using CSS
CSS is fundamentally based on the idea of economy: the more global and the less specific the better.
What this means is that if you're going to be using, say, the "Verdana" font everywhere on your site,
- Proper CSS would call for defining "Verdana" only once as the font-family for everything in a single, global declaration. No need to mention it anywhere else, since this global declaration takes care of it all.
- Very bad CSS would call for redefining endless times "Verdana" as the font-family in each and every class you create. This not only means puke-inducingly redundant declarations, but also means that if you change your mind about the font used on your site, you'll have to go through every single darned class you've created and modify it. That's just about as far as you can get from the point of using CSS in the first place.
If you're going to use a property "everywhere, but", define a property in a global declaration (probably in the <body> tag or such an overarching location), then make amendments at very specific locations where the rules don't apply. This will dramatically reduce/simplify/tidy your CSS.
On the other hand, if you're going to use a property at a decidedly very specific location, don't define it at a global location, as it will affect every other use of the related tag/class. For example, I cannot see, for the life of me, any good reason to globally add a right padding to every friggin' link on the site if all you want to do is separate items in your footer. By doing this, you'll risk looking perplexed in front of 20 people when you won't understand why your thumbnails don't line up in their cells. Chances are that most links on your site are not going to benefit from a seemingly arbitrary padding on the right, even if some very specific ones might.
While I'm on the topic of padding a link... "Padding" is space inside an element, while "margin" is space outside an element. What this means, in the case of a link, is that even if the two properties would have the seemingly visually identical effect, the padding would make the link itself larger, while the margin would make the space next to the link larger. That is important, because if you're using padding on a link (goodness forbid), the "padded" section of the link would be active (clickable) as well:
Notice how empty space is active (bad!)
That's just wrong. The appropriate property to define, in that case, would rather be a margin.
Redefining Links with CSS
Alright, you want to redefine links (the <a> tag) so that they all look the same, except for the "hover" variant. So you're going to spend a couple of minutes duplicating your half-dozen properties four times in all of these variants so that they are all the same, except for the "hover" which will have a slight difference.
Since you've been reading what I said about global versus specific, you know that's no good. Instead of duplicating every-compound-thing from a:link to a:visited and to a:hover and to a:active (phew!), just redefine the <a> tag itself, once! Boom: all the links will share a set of properties defined globally. Then add only the differences you want to the "hover" variant. You're done. That takes less time and is more economic. Especially if you want to make a slight change afterwards — you won't have to make that change four times.
Lastly, be careful when removing the underline from links. You may want to do it for aesthetics reasons, but good design calls for good usability, not just cuteness. Make sure, if you remove the underline, that links are still obviously links. The underline is such an established standard that you should think twice before you do that — or just do it where it doesn't break usability, such as in a menu where items are obviously selectable, while leaving the underline for links elsewhere in the text. Also, don't use underline on text that is not a link (used purposely here) — that's very misleading.