Why the building falls over
Vertical lines in the world stay parallel in a photograph only when the sensor plane is parallel to them. Tilt the camera up and the top of the building is farther from the sensor than the bottom, so it is rendered smaller, and the two vertical edges converge. It is the same geometry that makes railway tracks meet at the horizon, applied to a wall. The effect scales with how much you tilt and how wide you are, which is why the problem is worst in exactly the situation that causes it — a tall building on a narrow street where you cannot back up.
What a shift lens actually does
A perspective control lens is built with an image circle much larger than the sensor needs, and the shift mechanism slides the sensor’s window upward within that circle. The camera stays level, so verticals stay parallel, and you simply see more of the top of the building. Nothing is stretched or interpolated because every pixel was recorded optically. The tilt movement is a separate function that changes the plane of focus and is mostly useful for tabletop and product work rather than for buildings. For facade documentation, shift is the part that matters and 17mm and 24mm are the two focal lengths that cover almost everything.
The cost of fixing it in software
Correcting convergence in post is a keystone transform: the software stretches the top of the frame outward until the edges are parallel, then crops away what falls outside the rectangle. Three things happen. You lose resolution, typically 15 to 30 percent of the frame on a moderate correction. You lose field of view at the top, which is usually the part you were trying to include. And you interpolate pixels, which softens fine detail — mortar joints, shingle edges, grille louvers — in precisely the areas a survey frame exists to record. On a 45-megapixel body a mild correction is invisible. On a 24-megapixel body with a strong correction, a reader can tell.
How I decide on the day
If the frame is documentation and detail matters, the shift lens comes out. If the frame is context or marketing and I need to move fast, I shoot level with a wider lens than I need, leave headroom above the building, and correct later — shooting level and cropping down is a third option people forget, and it costs resolution without interpolating anything. If the building is tall and the street is narrow, I look for elevation instead of lenses: a second-floor window across the street, a parking structure, a lift already on site. Getting the camera to mid-height of the building is the real solution, and everything else is compensation for not being able to.
Level is a measurement
Turn on the electronic level and use it, because a camera that is one degree off vertical produces a wall that is visibly off and you will not see it on the rear screen in sunlight. Level in two axes, not one; roll is as visible as pitch. On a tripod I level the legs before the head so the pan stays true when I rotate for a second frame. And for any elevation that will be compared over time, note the camera height and the standoff distance, because reproducing that frame next year matters more than making it perfect today. Planning which hour to shoot the elevation is a separate question that has its own answer depending on which way the wall faces, which is the subject of planning a day around sun angle.
Frequently asked questions
Is a shift lens worth it for occasional architectural work? Renting is usually the right answer below about a dozen facade jobs a year. Buying makes sense once documentation is a regular part of the work.
Can a phone do perspective correction well? Phone software does it automatically and aggressively, often without telling you. Turn it off for documentation frames.
Should interior verticals always be perfectly parallel? For architectural and documentation work, yes. Deliberate tilt reads as a stylistic choice and undermines a technical set.