Simple formula for poor pictures: put a building in the shot, then don't aim
level. You end up with what is called "converging lines." The left and right side
of the building seem to meet somewhere above the picture. If that's what you
wanted, fine. If not, get it under control.
Level off. You need to have the lens (and film plane) parallel to the edges
that you don't want converging. (You don't actually have to be parallel to a side
of the building, just an edge.) For most buildings this means vertical. You
probably have zero height compared to a building, so for practical purposes this
means aiming at the base of the building. But, when you do this, you are probably
cutting off the top. So, you will back up quite a ways to get it to fit in. This
makes it small.
You'll notice now that the building is only using the top half of the film. My
example image shows beautiful green meadow, but you've got parking lots and hot
dog stands. At the very least, you will have to crop half the shot away, and
enlarge your grain twice as much as needed. Is there a better way?
Before looking at the real answer to this stumper, look at the numbers in
image 2. "35mm" film has images 36mm wide by 24mm tall. The "centerline" of each
shot is exactly half that: 12mm down for horizontal compositions and 18mm down
for vertical compositions.
Shift lenses do a few tricks but perhaps the main professional gimmick is
letting you control that centerline. Within limits, you can put it where you want
it, without changing the angle of your camera (and screwing up the
Shift lens documentation should tell you how much they shift. Let's look at
the Canon TS-E 24mm/3.5L. Like the other TS-E lenses, it shifts up to 11mm. This
is the distance the lens elements move up, but that also means the image moves
exactly that much on the film. Since film images are upside-down, moving the lens
up moves the image down. This means... fewer parking lots and hot dog stands. And
you can go back to your original vantage point and get a bigger image.
How far from your target do you need to be to get the desired shot without
twisting the perspective? Believe it or not, no trigonometry is needed.
It depends on the height of the target above your eye level, the height of the
film above the centerline (red, above), and the lens' focal length.
Using that, solve for anything if you know the other three. In practice, you
know the focal length and height-above-centerline numbers exactly, or can dial in
the height-above-centerline you need with a shift lens. Distance may be harder to
estimate, and height can be tricky. With buildings, you can usually guess 5
meters or 15 feet per story for commercial structures.
For now, this can help you picture what shots a given shift lens lets you
take. For instance, since the height-above-centerline is 12mm for a normal lens
and twice that (well, 23mm) for the TS-E lenses with maximum shift... we can get
twice as close to our target. Or, shoot a target twice as tall. Or use twice the
focal length (say, the TS-E 45mm instead of a normal EF 24mm).
Later, on a shoot, this can avoid a certain amount of trial-and-error. For
instance, say you have a 24mm lens with 11mm of shift, and a four story building
on a city street. You can back up to 65 feet away from your 60 foot target. The
formula says 65/60 = 24/X. X is about 22mm. You know that's right at the limit of
your shifting and the corners may darken somewhat. On the other hand, you not to
even bother getting your 1.4x TC out unless a vertical composition would work
(vertically, 65/60 = 36/X, X would be 33.3mm, and a you could get a maximum
height-above-centerline of 18mm + 11mm x 1.4, or 33.4mm).
When maximum shift is employed, shift lenses (with their unnaturally wide
field of view) can show some light falloff in the corners, due to basic optical
properties. In addition, even the widest filters may be visible in the corners of
the shot. Smaller apertures typically fix the first problem but accentuate the
second. Hopefully you have the option of backing away from your target and
reducing shift slightly. If not, consider using an auxiliary filter holder
instead of screw-on filters.
There are not yet any auto-focus shift lenses, even the Canon TS-E lenses with
the auto-focus mount. Some shift lenses are a bit more primitive, making you stop
down manually before shooting. This will be true for any "old" mount with manual
aperture control, because it would take Rube Goldberg hardware to translate the
mechanical movement at the mount to movement at the aperture, when half the lens
moves up and down.
You will be paying quite a bit extra, both for the optics that make a larger,
shift-able image, and the hardware on the lens that let you move the glass up and
down. Even then, the lens will probably have a smaller maximum aperture than
you'd expect. For instance, a Canon 24mm/2.8 is about $500 while the shifting
24mm/3.5 is more like $1200 on the street. If its any consolation, the makers are
probably losing money on every one they make due to the low volume of production.
You'll see all three Canon TS-E lenses for the same price, even with their vastly
different optics, because Canon is offering them to complete its lineup, and not
attempting to price them to recoup their cost.
A tele-converter (TC) is a lens group that attaches between your objective
lens and camera body. They typically magnify the image size by a factor of 1.4x,
1.5x, 2x, or 3x. TC's also (sadly) increase your f-stop by the same factor.
Generally any older TC or 2x TC with less than seven elements will be crap.
On shift lenses, a TC will also (for practical purposes) increase the amount
of your shift. The Canon TS-E lenses go from 11mm shift to 15.4mm (1.4x) or 22mm
shift (2x). Looking at image #3, guess where you can now place your ground level.
The horizontal format can suddenly lose the bottom third of your building, when
the centerline moves to 10mm below the bottom of the frame with a 2x. The
vertical format can move the centerline to 4mm below too. (Just as lens specs are
not exact, teleconverter specs can be pretty loose. A nominally 2x converter may
be 1.8x or 1.9x. Never seem to be better than advertised, even for Leitz and
Now, compare the Canon TS-E 45mm/2.8 vs. the TS-E 90mm/2.8. With a 2x TC, the
45mm/2.8 turns into a 90mm/5.6... but with 22mm shift possible instead of 11mm.
Which sounds like more fun?
Since TC's in effect take their picture from the center of the objective lens'
image, they avoid problems that tend to arise in the corners of the
But Wait! That's not All!
Any Shot with Lines!
The example buildings can be replaced with trees. Show those forest canopies
while avoiding the "acid trip trees" look.
Or any other shot where there are two parallel lines... and if that sounds like
"most shots" you get the picture. Back before rolls of film were invented, there
was a time when all camera lenses shifted, and truing perspective was considered
as important as focus and selecting shutter speed and aperture. (Back then, they
also used a variety of terms for shift: rise was shift up, etc. Archaic people
still do so today 8-)
With the first professional-use "box" cameras (as opposed to bellows cameras),
a generation of photographers - and photograph viewers - grew up not having this
As time went on, all the serious 35mm camera marques issued "perspective
correction" lenses. Or perhaps I should say, lens, usually with a focal length of
35mm. Zeiss does a 35mm for Contax, Leitz does a 35mm for Leica. Olympus and
Pentax have one too. Minolta made a 35mm for the old manual lens mount, but
doesn't have one for the auto-focus cameras. Nikon made (and makes) the 28mm and
35mm. Canon had the TS 24mm and 35mm for the manual mount, and the TS-E 24mm,
45mm, and 90mm for the current electronic mount.
Besides fixing perspective, you can use the same feature to screw up normal
perspective. I took a band's CD cover picture, where the guitars are the size of
battleships and the guitarists' heads are the size of gnats. Art is Truth.
Look at the example vertical image #3. Assuming your shift lens shifts left
and right (Canon TS-E's do), you can take the left half and right half of a
panorama on separate frames, and combine the images in the darkroom or a photo
touch-up program. There will be some overlap to help you assemble the halves into
the final image. This overlap can be nominal if you use a TC.
You can end up with a scene that uses about 70mm width and 24mm height. This
will give you strictly higher image quality that you'd get with a 6x6 medium
format camera and cropping to get a panoramic slice - merely 55mm x 20mm. If
anyone cares, that’s 1.5x larger negative... advantage 35mm.
The Canon TS-E lenses will shift in any direction, and have clicks every 30
degrees. You can thus also get a shift somewhat up or down at the same time you
shift hard left to get one half, and reproduce the exact shift amount and angle
to take the second half.
Normally you can change your viewing angle to select the foreground and
background that surround your actual subject. You may have artistic conflict,
however, if there is a strong rectangular element in the subject that you want to
render squarely. Examples are mirrors, railings, and windows.
If you face a mirror head on, you will be in the shot. If you move to the
side, it turns into a parallelogram. With a shift lens mounted, move the camera
to the side, turn the film and lens parallel to the mirror, then use shift to get
the original composition back (more or less).
Sometimes you have limited access to a fixed display, especially in tight
interiors. There might be a ceiling support exactly where you'd like to stand.
Set up to one side and shift horizontally to reframe.
Depth of Field!
Besides shifting in any direction, the Canon TS-E lenses also tilt. Actually,
this has nothing to do with the above shift stuff, except that the only lenses
that tilt are the Canon TS-E's. What does tilt do for you?
Normally, everything in a plane a set distance from, and perpendicular to your
lens will be in focus. When you use lens tilt, you forego the "perpendicular"
constraint. You can tilt that plane quite severely if desired. The plane of the
back element of the lens, the film plane, and the focus plane will all
The classic example is for landscapes. You have wildflowers at your feet and
Alps at infinity. f/22 won't give you the depth of field you need (well, it would
at 24mm but at the sacrifice of sharpness). As its just after sunset there's not
much light, and there is also a wind moving the flowers so a long exposure won't
work. Fast shutter dictates big aperture, which means lousy depth-of-field...
As you tilt the lens, you make the focus plane tilt. However, a small tilt of
the lens creates a huge tilt in the focus plane. Just 10 degrees of lens tilt can
rotate the focus plane to the point that wild flowers a meter away, at the bottom
of the frame, are perfectly focused as are distant alps.
There are very mathematical explanations to calculate all the exact effect,
but I don't know how you could use them accurately in the field. It would be nice
if there was a focusing screen with a half-prism focusing aid at both the top and
bottom of the frame. But there's not. The camera could also calculate this
information for you and display it.
The other tilt trick is just the opposite. Say you want everything out of
focus except for the subject. Say there is a row of columns ahead of you, running
left to right. Tilt like mad, focus one column, and the equally distant neighbors
to the left and right will be fuzzy.
Why You Can't Get a Good Exposure
Wide angle lenses get darker the further from center you get. This is called
"cos^4 vignetting" and is due to the light hitting the film at an angle. This is
a problem when the lens->film distance is small compared to the image
On SLR's, wide angle cameras use a "reverse telephoto" group at the back, so
the rear of the lens is far enough away to leave room for the mirror. This extra
glass hurts quality, but means that even a EF14mm/2.8L is about 40mm from the
film, which is 36mm wide... which means "cos^4" is NOT visible on SLR
wide-angles. Rangefinders like Leica M and Contax G and Mamiya 7 have no
telephoto group, so their images are sharper - but the back of the lens is VERY
close to the film, and the corners are much darker.
So what does this have to do with your shift lens? Everything! Shift lenses
make a picture much wider than a normal lens, but the distance from lens to film
is still about 40mm. Your lens shifts 11mm, so the image is 36mm (film) + 11mm *
2 = 58mm or so. Unshifted, only the center 24mmx36mm of the image hits the film,
and has no more cos^4 vignetting than a normal 35mm. However, as you shift, the
darker part of the total image hits the film.
If you externally meter, then a meter reading good for a centered shot will be
too short an exposure for a shifted shot. How much too short? Who knows, bracket!
8-) On the other hand, the camera meter depends on taking readings of light off
of partial reflections, and depend on the angle the light hits the sensors. When
you shift, the light comes from a different direction and the sensors may see
none of it (or too much), so they will suggest a bad exposure.
How to Get a Good Exposure
If you don't want to bracket, you need to build up some base-line cos^4 data
for your lens. Perhaps the following might be agreeable:
Build comparison series: on slide film, photograph a grey card metered with 0
shift and no exposure compensation, then -1/3, -2/3, -1, -1 1/3 stop. (either
internal meter or external is fine)
Find worst-case dropoff: shift maximum (horizontally if you can shift
sideways), and use the same shutter and aperture as the first shot of your
Develop these shots.
Using the center of the series shots, determine whether the maximum shift is
-2/3 darker or whatever. That maximum shift is 36mm(width)/2+11mm(shift)= 29mm
from image center. If you want, measure the falloff at other distances.
Then: after metering a scene with a handheld meter or unshifted lens, set the
exposure manually but with the test-result compensation. This should be perfectly
accurate with one caveat: Because the darkening is progressive, the
maximally-shifted side will always be somewhat darker. For arrchitecture, this is
perfect - it saves you from using an actual graduated filter to bring the sky and
ground brightness closer.
Since you may not be able to see this difference in the viewfinder, your
probably still want to bracket, but bracket from the external/unshifted exposure
to your test-result maximum compensation. Some picture in that range should be
The only return I get on writing an article like this is the right to voice my
own opinions - and hopefully Canon's listening!
More than anything in the world, I want a 35mm tilt-shift lens. The gap
between the 24mm and 45mm is just too great. And, the 24mm is not sharp enough to
have its images cropped 50% without pain. This lens physically works with the
Extender EF 1.4x teleconverter (TC), but resolution seems to suffer somewhat. In
effect, the TC creates a very expensive, 35mm f/4.9 lens with less than ideal
sharpness. Instead, a one-piece TS-E 35mm F/2.8 or 3.5 lens should have much
better optical quality (no TC), as well as being cheaper (no TC), lighter (no
TC), and brighter (no TC).
It would be nice if the TS-E lenses became TS-EF - autofocus. I firmly believe
that everyone's 24mm (and 35mm...) lens should have tilt and shift, for when they
are needed, but autofocus when they aren't. Canon, with their great
all-electronic mount, could easily allow this.
New Focusing Screens
We need a new focusing screen for shifted composition. Canon makes a screen
with a grid, for verifying perspective. This is great, except the TS-E lenses are
all manual focus, and the grid screen doesn't have a focus aid. There is a
perfectly good screen with the usual split-circle, but no grid lines of any sort.
So, we are left estimating ranges and looking at the distance scale on the focus
For tilted composition, the situation is even worse - the distance scale only
tells us what is happening at the image center. The scale might say 3 meters when
the lens is actually focused at infinity on one edge and 50cm at the other.
Calculating lens tilt mathematically is practically a black art. I only know a
couple people that even know how, and they don't do it in their head on the
scene. They guess a tilt, then squint at the viewfinder trying to guess if it is
sharp enough. This is hard at f/3.5 (the Canon 24mm lens) or f/4.9 (same with
1.4x TC). I'd like a focusing screen with the normal split-circle focus aid, but
instead of merely the image center, have a circle (or major portion thereof) at
each corner or perhaps centered on each of four sides.
More Advanced Camera Electronics
An second alternative would be a camera that tells us what focus we're at, at
the extreme edges of the camera. This would involve the camera asking the lens 1)
focus distance, 2) tilt, 3) rotation of tilt, and 4) a constant or two related to
focal length. The camera then calculates the focus distance extremes and displays
them on the LCD. A photographer who could estimate ranges accurately could adjust
the lens until the readout matches his estimates.
A third alternative would have two focusing rings for near and far focus. It
would use Canon "E-M". (This means that the focusing ring is just an electronic
input sensed by the computer, that drives a motor to change focus.) We would get
a focus motor, and a second motor that tilts the lens appropriately. The focus
rings would be right next to each other, so normal operation would involve
grabbing both simultaneously.
If this two-ring lens had autofocus (in untilted mode), Canon wouldn't mark
distances on the focus rings, making the split focus screen mandatory. If the
lens was manual focus only, the rings would have full markings, and ranges would
be estimated by the photographer. I don't even postulate a camera with AF sensors
at the frame's extreme corners, that attempted to tilt, rotate, and focus,
Adding electric control of the current tilt probably would be awful.
Hopefully, just as many lenses are now internal or rear focus, the tilt effect
could be accomplished simply by tilting one internal group.
I hope at this point, you are sadder but wiser. Ideally, you should be
discontent with every non-shift (and non-tilt) lens you have. This consumer
revolt will get back to Canon et al, who will finally start making the lenses the
world really needs.
[Editor's note: You might have trouble finding a tilt-shift lens at your
neighborhood or shopping mall camera store. Try one of
This article is copyright 1997 Frank Sheeran. If you want
to reuse it in whole or portion, contact