The horizon tells you where the camera stood

One line that solves most placement problems

In any photograph taken with the camera level, the horizon line sits at the camera’s height. Not the middle of the frame: the camera’s height above the ground.

That gives you a free measurement. Find something in the plate whose real height you can guess: a person, a door, a chair, a car. Where the horizon crosses them tells you how high the camera was. If the horizon crosses a standing adult at chest level, the camera was at chest height, roughly 1.4 metres. If it crosses at their knees, the camera was low.

And it works the other way. Every person standing on the same ground plane has the horizon crossing them at the same point on their body. Two people at different distances, one large and one small in the frame, both have the horizon at their chest. This is the single most useful rule in compositing people into scenes, and it is exact, not approximate.

So the placement procedure becomes:

1. Find the horizon in the plate. Even if it is hidden behind buildings, you can locate it: it is where receding parallel lines converge, and it is the eye level of anyone standing on the ground.
2. Find the same feature on the element. Where does the horizon cross the person in their original photograph?
3. Scale and position the element so the plate’s horizon crosses it at the same point.

That gets the scale right without any guesswork about metres.

When the cameras were at different heights

If the element was shot from above and the plate from eye level, the horizon rule tells you they do not match, and no scaling fixes it: you would be looking down at the person and along at the street.

The fixes, in order of quality:

• Reshoot the element at the right height.
• Use it where the mismatch does not show: put the person further back, where the perspective difference is smaller, or crop to a region where the angle is ambiguous.
• Perspective-warp the element. This works for small differences on simple shapes and produces distortion on anything complex, because a warp is a 2D operation being asked to simulate a 3D viewpoint change.
• Accept it and make the composite graphic rather than photographic.

Focal length and the shape of things

Focal length does not change perspective, distance does, but the two travel together, because a wide lens is used close and a long lens far away.

What that means for compositing:

• A wide lens used close exaggerates the difference in size between near and far parts of the subject. A face photographed at 24 mm from half a metre has a large nose and small ears.
• A long lens from far away compresses those differences. Backgrounds look larger and closer relative to the subject.

So an element shot at 85 mm dropped into a plate shot at 24 mm sits oddly, because the internal proportions of the element do not match the perspective of the scene around it. Check the EXIF of both files. A mismatch of 35 mm against 50 mm is usually survivable; 24 mm against 135 mm is not.

Grounding: the contact point decides everything

An object’s position in depth is read from where it meets the ground, not from its size. Put a cut-out chair with its feet high in the frame and it reads as far away, regardless of how large it is; the eye then judges its size relative to that distance and concludes it is enormous.

This is why the base of an object has to sit exactly on the correct ground line, and why a floating element is not merely missing a shadow: it has no position at all.

For an object that should be partly behind something, occlusion beats everything: masking the element so a table edge or a plant passes in front of it does more for believability than any amount of light matching, because occlusion is unambiguous depth information.

Converging lines and the ground plane

When you need to place something on a receding surface (a poster on a wall at an angle, a rug on a floor, a screen on a laptop) the element has to follow the plane’s perspective.

The method:

1. Identify two pairs of lines in the plate that are parallel in reality: the top and bottom of a wall, the two sides of a floor.
2. Extend each pair until they meet. That meeting point is the vanishing point for that direction.
3. Transform the element so its own edges point at the same vanishing points.

Photoshop’s Vanishing Point filter lets you define the plane and then paste into it, doing the transform automatically. GIMP’s Perspective transform with the grid overlay does it manually, and the free tool Hugin handles extreme cases. In any tool, turning on a grid or dragging guides along the plate’s real lines makes this a two-minute job instead of a fiddly one.

Checking

Three tests, all fast:

Extend the lines. Draw a temporary line along the base of the element and along a real edge in the plate. Do they converge to the same point?

Compare heads or tops. For multiple people or objects on the same plane, the line through their tops and the line through their bases should converge at a single point on the horizon.

Look at the element’s own ground contact at 100 per cent. A foot half a pixel off the floor is a visible float; a foot overlapping the floor by two pixels reads as sunk into it.

The limit

A photograph is a projection of three dimensions onto two, and the projection is not invertible. You cannot recover the camera’s exact position, and you cannot change a photograph’s viewpoint after the fact: a warp fakes it convincingly only for small angles on flat or simple surfaces.

Where the two photographs’ viewpoints disagree, the honest answer is that they cannot be combined convincingly, and the work should be redirected: reshoot one, choose a different plate, or design the image so the mismatch is not a claim about a real scene.