The horizon is at eye level — if the picture plane is vertical

Of everything in this subject, one fact does the most work for the least effort: the horizon in a picture is at the height of the eye that made it, and it cuts every object of a given height at the same fraction of that height, however far away the object is.

It means a figure standing on the ground can be drawn at any distance without any construction. Find where the horizon crosses the near figure, note the fraction, and every other figure of the same height is crossed at the same fraction. No vanishing points, no measuring, no plan.

It is also stated without its condition almost everywhere it appears, and the condition is not a technicality.

Why the horizon is at eye level #

The horizon is the image of the ground plane’s line at infinity: the images of the points infinitely far away across the ground.

A ray from the eye to a point infinitely far away across a level plane is a horizontal ray. Horizontal rays leave the eye at the eye’s own height, and they stay at it. So the images of all such points lie where horizontal rays meet the picture plane, which is at the eye’s height in the picture.

That is the whole derivation. Two consequences fall out immediately.

The horizon does not move when the viewer walks about, provided their eye stays at the same height. Points at infinity have direction and no position, so translating the eye horizontally does not change any of them. Climbing a step moves the horizon; walking a mile does not.

Anything at the eye’s height, anywhere in the scene, is on the horizon. Not near it — on it. The eyes of a crowd of people all the same height as the viewer lie exactly along the horizon, at every distance, which is a striking thing to notice in a photograph and is diagnostic: it locates the camera’s height in a picture with no other information.

The fraction, and why it is constant #

Now the useful version. A vertical object of height h stands on the ground. Where does the horizon cross it, as a fraction of its drawn height?

With a vertical picture plane, the answer is the eye height divided by h, at every distance.

The reason is that with the picture plane vertical, the depth of a point on a vertical line does not change as one moves up the line — every point on that vertical is at the same distance from the picture plane. So the image height is an affine function of the world height: divide by a fixed depth, and nothing else. Affine maps preserve ratios, so the point at world height 1.62 m images at 1.62/1.78 of the way up the drawn figure, whatever the depth happens to be.

Measured across four figures at four distances, the four fractions agree to better than 10⁻⁹, which is the arithmetic. The rule is exact, not approximate.

What breaks when the camera tilts #

Tilt the camera — aim it down at the ground, or up at a building — and the picture plane is no longer vertical. Now the depth of a point on a vertical world line does change as one moves up the line, because the line is no longer parallel to the picture plane.

The map from world height to image height stops being affine and becomes projective: a ratio of two linear functions. Projective maps do not preserve ratios, which is the whole subject of this site, so the fraction is no longer the same for figures at different depths.

At twelve degrees of tilt the four fractions in the figure spread by 1.11 percentage points. That does not sound like much, and on a 200-pixel figure it is two pixels of misplacement at one distance relative to another — enough to make a group of figures sit oddly on the ground without any single one looking wrong.

The rule, stated properly, is: with a vertical picture plane, the horizon cuts every object of a given height at the same fraction. The condition is the interesting half, because it explains when the shortcut can be used and what to do when it cannot.

The same condition, three ways #

It is worth noticing that a single geometric condition is behind three separate rules that are usually taught apart.

Verticals stay vertical when the picture plane is vertical, because then the vertical world direction is parallel to it and has no finite vanishing point. That is the definition of the two-point case.

The horizon cuts equal heights at equal fractions when the picture plane is vertical, for the affine reason above.

A photograph of a building does not lean when the picture plane is vertical, which is why architectural photographers shift the lens instead of tilting the camera.

All three are the same condition. A drawer who knows that the three go together has one thing to remember instead of three, and knows immediately that a drawing with converging verticals cannot use the eye-level fraction shortcut.

Using it #

The rule is the fastest way there is to place figures in a drawing, and the procedure is short.

Draw the horizon at the height the viewer’s eye is meant to be. Draw one figure anywhere, at whatever size suits the composition. Note where the horizon crosses it. Every other figure standing on the same ground, of the same height, is crossed at the same fraction — so its size at any position follows from where its feet are placed.

For a figure whose feet are placed, the drawn height is fixed. For a figure whose height is chosen, the feet position is fixed. Either way one decision determines the other, and no construction is needed.

Three refinements are worth having.

Figures of a different height are handled by scaling: a child of 1.2 m is crossed at 1.62/1.2, which is greater than 1, meaning the horizon passes above the child’s head. That is correct and is what makes a crowd of adults and children read properly.

Figures not standing on the ground — on a step, on a balcony — need the height of the surface they stand on added, which requires an actual construction. The shortcut only covers the common case.

A camera at an unusual height changes everything at once and is the fastest way to change the feel of a drawing. An eye at 0.7 m makes a room look enormous and is how a child’s-eye view is drawn; an eye at 2.5 m produces the slightly detached quality of an architectural view.

What the rule does not say #

Three misreadings are worth heading off, because each produces a plausible drawing that is wrong.

It does not say the horizon is in the middle of the picture. The horizon is at eye level, and where eye level falls in the frame depends on how the camera is aimed within the vertical picture plane — or, in a drawing, on where the drawer puts it. A low horizon and a high horizon are both correct and imply different framings.

It does not say objects shrink at a rate the rule gives. The rule fixes where the horizon crosses a figure. It says nothing about how the figure’s drawn height changes with distance, which depends on depth and is what the measuring-point construction is for. The two are complementary: the horizon rule links height to feet position, and the depth construction says where the feet go.

It does not say the horizon is where the sky meets the ground. In flat open country it is, near enough. In a room there is no visible horizon at all, and the horizon of the drawing is still exactly as real and exactly as useful — it is a line in the picture, not a feature of the landscape. Hills, buildings and trees rise above it and none of that moves it.

The reason it is worth a whole essay #

Because it is the one place where the geometry of this subject pays out immediately, in a form usable without any calculation, and because it is nearly always given without the condition that makes it true.

A rule with an unstated condition is worse than a rule with a stated one, even when the condition holds most of the time. It holds for level cameras, which is most drawings and most photographs, so it works, and then it silently stops working on the drawing of the tall building — which is exactly the drawing where getting the figures right matters and where the drawer has no reason to suspect the shortcut has expired.

Measuring the failure rather than describing it puts a number on when to stop trusting it: 1.11 percentage points of spread at twelve degrees, which is visible, and proportionally less below that. Under about five degrees of tilt the rule is still good enough for figure placement; above it, the construction is needed.

What a painter does with the horizon’s height #

The rule fixes a relationship; where to put the horizon in the frame is a free choice, and it is one of the few decisions in perspective that is genuinely compositional rather than geometric.

A low horizon puts the eye near the ground. Everything above eye level is seen from below, figures loom, and the sky takes most of the frame. It is the view of a child or of someone lying down, and it makes ordinary objects monumental. Dutch landscape painting uses it to give the sky four-fifths of the canvas.

A high horizon puts the eye above most of the scene. The ground plane opens out, spatial relationships between things on it become legible, and figures read as arranged rather than as encountered. Maps and battle scenes want this, and so does any picture whose subject is a layout.

A horizon at the height of the principal figures’ eyes puts the viewer among them. Every figure of that height is crossed at the top, which reads as being addressed at eye level, and it is the standard choice for group portraits and for narrative painting where the viewer is meant to be present rather than looking on.

None of that is geometry. What the geometry contributes is that the choice is a single number, it applies to the whole picture at once, and once it is made every figure’s size is determined by where its feet go. A painter who moves the horizon has re-sized every figure in the composition, which is why it is decided first.

Reading the eye height out of a photograph #

The rule runs backwards, and doing so is the quickest piece of forensic geometry there is.

Find anything in the picture whose real height is known and which stands on the same ground as the camera — a person, a door, a car. See where the horizon crosses it as a fraction of its drawn height. Multiply the fraction by the real height. That is how high the camera was.

If the horizon is not visible, it can be found first: two families of horizontal parallel edges give two vanishing points, and the line through them is the horizon whether or not any sky appears.

The result is often informative. A photograph whose horizon crosses standing figures at chest height was taken from below the height of the people in it — from a crouch, or by a shorter photographer, or from a vehicle. One that crosses above their heads was taken from a step or a raised platform. Neither is visible as such until the horizon is located, and both change how the picture reads.

The same procedure applied to a painting says what viewpoint the painter chose, which is a different question from where they were standing, and is usually the more interesting one.

The condition, one more time #

The whole of the above assumes a vertical picture plane, and it is worth ending on how to tell whether that assumption holds in a picture that someone else made.

Look at the verticals. If the vertical edges of buildings, door frames and lamp posts are parallel to each other in the picture, the picture plane is vertical and everything in this essay applies. If they converge — toward a point above or below the frame — the camera was tilted, the fractions drift with depth, and the shortcut is not available.

That check takes a second and needs no measurement, which is a fair summary of why the condition is worth carrying: it is easy to test, it is usually satisfied, and the one case where it fails is the case where a drawer is most likely to need help and least likely to notice they have lost it.

The rule as a design instrument #

Because the horizon’s height fixes every figure’s size once its feet are placed, moving it is the fastest structural change available to a drawing, and it is worth listing what moves with it.

Raising the horizon lowers the apparent viewpoint relative to the scene — no, precisely the reverse: raising the horizon in the frame corresponds to raising the eye, which opens the ground plane out, spreads objects apart in depth, and makes the arrangement of things legible.

Lowering it drops the eye, closes the ground plane to a narrow band, stacks objects against each other, and makes the sky and the objects’ upper parts dominant.

Neither is more correct. What matters is that the choice is a single number applying to the whole picture at once, and that every figure in the composition is re-sized by it — which is why the horizon is decided before anything is drawn, and why moving it late is expensive.

The same number also fixes the relationship between the viewer and the subject. A horizon at the height of the principal figures’ eyes places the viewer among them; well below, and the viewer is looking up at them; well above, and the viewer is looking down. That reading is reliable because it is geometric: the horizon is the viewer’s eye height, and everything above it is above the viewer.

What to check in someone else’s picture #

Two checks, both quick, both diagnostic.

Are the verticals parallel? If yes, the picture plane is vertical, the rule applies, and every figure of a given height should be crossed at the same fraction. If they converge, the camera was tilted and the fractions drift — by about a percentage point at twelve degrees.

Do the figures obey it? Measure the crossing fraction on two figures at different distances. Equal means the picture is consistent; unequal, with parallel verticals, means the figures were placed by eye rather than constructed, which is the same class of fault as a depth row spaced by judgement.

Both take a straightedge and no arithmetic, and between them they say whether a picture’s figures are in the scene or on it.