One, two and three point are one construction

Books on perspective are usually organised around three chapters: one-point, two-point and three-point. They read as three techniques of increasing difficulty, to be learned in order, each with its own construction and its own set of rules.

They are one construction. The three names count how many of an object’s vanishing points land at a finite place in the picture, and that count is decided by how the object is turned relative to the picture plane and how the camera is aimed — not by any choice the drawer makes about method.

What is being counted #

A rectangular box has twelve edges in three families of four, each family parallel to one of three mutually perpendicular directions. Each family has a vanishing point: the image of that direction’s point at infinity.

Whether that vanishing point is somewhere on the paper depends on one thing. If the direction is parallel to the picture plane, its ray from the eye never meets the picture plane, and the image of its point at infinity is itself at infinity — the family’s edges stay parallel in the drawing. If the direction is not parallel to the picture plane, the vanishing point is at a finite place, and the edges converge on it.

So the three cases are:

One-point. Two of the three directions are parallel to the picture plane. The box is face-on and the camera is level. Verticals stay vertical, horizontals stay horizontal, and only the depth direction converges.

Two-point. One direction is parallel to the picture plane, and it is nearly always the vertical, because the camera is level. The two horizontal directions converge to two points on the horizon.

Three-point. No direction is parallel to the picture plane. The camera is tilted, so even the verticals converge, and the third vanishing point sits far above or below the frame.

The count is a property of the arrangement. A drawer who turns the box has changed which case they are in, whether or not they intended to.

The transition is continuous #

The most misleading thing about the three-chapter treatment is that it makes the cases sound discrete, as if a vanishing point were a thing that exists or does not.

Take a box face-on and rotate it slowly. At exactly zero the horizontal edges across the front are parallel to the picture plane and stay parallel in the drawing. At one degree they are not, and their vanishing point is at a finite place — a very long way off, thousands of picture-widths away, but finite. At ten degrees it is closer. At forty-five it is a comfortable distance to either side.

Nothing appeared. The vanishing point was always there; it came in from infinity as the box turned, continuously, and the picture at one degree is indistinguishable from the picture at zero because a vanishing point ten thousand pixels away produces convergence too slight to see.

That has a practical consequence. A drawing constructed as a one-point perspective of a box that is very slightly turned is wrong in a way that will never be visible, and a drawing constructed as a two-point perspective with the vanishing points placed conveniently on the page is wrong in a way that is very visible indeed, because putting both vanishing points on the sheet forces an extremely wide lens.

Both points on the sheet is a lens, and the number is exact #

The claim that putting both vanishing points on the page forces an extremely wide lens can be made a number, and the number is one nobody would choose.

Two perpendicular horizontal directions have vanishing points whose offsets from the centric point satisfy $x_1 x_2 = -f^{2}$ — the perpendicularity condition, and the same pole-and-polar relation the tilted plane obeys vertically. Requiring both to sit inside a sheet $W$ wide means $|x_1|, |x_2| \le W/2$, so

$$f^{2} \;=\; |x_1 x_2| \;\le\; \frac{W^{2}}{4}, \qquad\text{that is}\qquad \theta \;\ge\; 90°.$$

Both vanishing points on the paper is a field of view of ninety degrees or more, exactly, with the bound reached only when both sit precisely at the edges. Putting them comfortably inside — at a quarter of the width either side — gives $f = W/4$ and a field of 127°, which is wider than any rectilinear lens made.

Read as a viewing distance that is 0.5 picture widths at the bound and 0.25 inside it: a 160 mm drawing correct from 80 mm, or from 40. So the two-point drawing a student is taught to lay out on one sheet is, without anybody saying so, a picture that has to be read with the nose against it — which is the free parameter this field keeps finding, chosen here by the size of the paper.

The vertical is not special #

Two-point perspective is nearly always drawn with the verticals staying vertical, and the reason is worth stating because it is a convention rather than a law.

The verticals stay vertical when the picture plane is vertical, which means when the camera is level. That is how people photograph and how buildings are drawn, so it is the usual case. But it is the camera’s attitude doing the work, not anything about verticality.

Aim the camera down at a box from above and the vertical direction is no longer parallel to the picture plane. Its vanishing point comes in from below, and the box’s vertical edges converge downward. Aim up at a building and they converge upward, which is why photographs of buildings lean backwards whenever the camera is tilted to fit the top in.

The professional response is not to correct it afterwards but to avoid producing it: keep the picture plane vertical and shift the lens up instead, which moves the principal point off the frame’s centre without tilting anything. That is what an architectural shift lens does, and it is a piece of hardware that exists entirely to keep a drawing in the two-point case.

What actually changes, and what does not #

Across all three cases, the following are unchanged:

The horizon is still the image of the ground plane’s line at infinity, and every horizontal direction still has its vanishing point on it. In the three-point case the horizon may be off the frame, which does not make it absent.

Every family of parallel edges still converges to one point, and that point is still findable from the drawn edges by least squares, with the residual saying whether they really share it.

The cross-ratio is still the invariant, and the measuring-point construction still works unchanged.

The relation that recovers the focal length still holds for every pair of perpendicular directions — with the caveat that in the one- and two-point cases one of the directions has its vanishing point at infinity, so that pair supplies no equation and the recovery needs the principal point supplied rather than derived.

What changes is only the count of finite vanishing points, and therefore how much of the geometry is visible on the paper. Which makes the three-chapter organisation of the textbooks an organisation by convenience of drawing, not by kind.

Why one-point survived as a teaching device #

There is a real reason one-point perspective is taught first, and it is not that it is a simpler case of a general method.

It is that in the one-point case the picture plane is parallel to the front face of the object, so that face is reproduced to scale. Its proportions are exact, its right angles are right angles, and its dimensions can be measured off the drawing with a ruler. Only the depth direction is foreshortened.

That makes one-point perspective a hybrid: an elevation for everything in the frontal plane and a perspective for everything receding. It is why room interiors and stage sets are drawn this way — the wall that faces the viewer can be laid out from the plan directly — and it is the same reason oblique projections keep one face undistorted and are used for the same kind of work.

The disadvantage arrives with the depth. Everything about how far back things are sits in one direction, and getting it right requires the measuring-point construction, which is exactly the part most treatments hurry past.

Where the two-point construction goes wrong #

The two-point case is the one that gets drawn most and the one with the free step.

The method places two vanishing points, draws the near vertical edge, runs the top and bottom edges to the vanishing points, and then places the two far vertical edges. That last placement is judgement. No printed method supplies a construction for it, because doing it properly requires a measuring point that the method never introduces.

The drawn result is a picture of a box. Which box is decided by that step, and the recovery says which: placing the two far edges symmetrically gives a cube for free, and eight points of asymmetry gives a box 1.4× shallower than it is wide, with nothing in the picture to indicate the difference.

That is not an argument for abandoning the construction. It is an argument for knowing that the construction has an unconstrained parameter in it, and for using the measuring point when the proportions matter.

The count for objects that are not boxes #

The vocabulary of one, two and three points is about boxes, and it stops making sense as soon as the object is not one.

A hexagonal prism has three horizontal edge directions plus the vertical, so four vanishing points, three of them on the horizon. A staircase has the directions of the treads, the risers and the string, and the string’s direction is neither horizontal nor vertical, so its vanishing point is off the horizon and not at the vertical point either — which is the detail that makes stairs hard to draw and easy to check.

A cylinder has one axis direction with a vanishing point and a whole family of surface lines that all share it. A sphere has none. A curved road has a vanishing point for each of its tangent directions, and they walk along the horizon as the road turns.

So the honest general statement is: every direction in the scene has a vanishing point, most of them are irrelevant to any particular drawing, and the number that matters is however many the object supplies. “Three-point perspective” names the case of a box with a tilted camera, and there is no such thing as four-point perspective — there is a hexagonal prism.

What to do with this #

The practical upshot is small and freeing.

Stop choosing a “type” of perspective before starting. Decide where the eye is, what it is aimed at, how wide the view is, and how the object is turned. Those four decisions fix everything, including how many vanishing points will be finite and where they will be, and the answer may well be that both horizontal ones are off the paper — which is the normal condition of a photograph and is not a problem to be avoided.

If the vanishing points are off the paper, the constructions that need them can be replaced by ones that do not. The measuring-point method works with a vanishing point anywhere, including off the sheet, provided the direction to it is known. And where a construction genuinely needs a point that is metres away from the drawing board, that is a signal worth heeding: it means the drawing is a long-lens view, which is usually what was wanted and rarely what gets drawn.

The history, which explains the chapter structure #

The three-chapter organisation is not arbitrary. It is the order in which the cases were worked out, and each step took a long time.

Brunelleschi’s demonstration around 1425 and Alberti’s account of it in 1435 handle the one-point case: a picture plane parallel to the principal face, a single centric point, and a construction for the depths. Alberti’s costruzione legittima is essentially the measuring-point method with the eye’s distance from the picture plane made explicit as a separate drawing off to the side.

The two-point case took another century to be treated properly, and it is genuinely harder to construct by hand: two vanishing points, two measuring points, and the depths of both horizontal directions to be laid out. Vignola and Barbaro have it by the 1560s.

The three-point case is later still, and it is telling that it was of no interest to painters. A painting hangs on a wall and is looked at level; there is no reason to tilt the picture plane. Three-point perspective becomes useful when the subject is a tall building seen from the street or from above, which is to say when photography and modern architecture arrive together, and it enters the manuals in the twentieth century.

So the chapters are a history, and reading them as a hierarchy of difficulty is a misreading of what the difficulty was. The hard part was never the concept of the third vanishing point; it was constructing any of it with a straightedge before anyone had coordinates.

Where the modern version of this lives #

Every three-dimensional graphics system in use does exactly what this site’s camera.js does: a projection matrix built from a field of view, an aspect ratio and a near plane, applied to vertices, with the perspective divide at the end.

None of them has a concept of one-, two- or three-point perspective, and none of them needs one. The count of finite vanishing points is an emergent property of where the camera is pointing, exactly as it is here, and nobody using a game engine or a CAD system ever selects it.

That is the strongest evidence that the three-chapter division is about hand construction rather than about geometry. The moment the projection is computed instead of constructed, the distinction stops being available to make — which is also why a drawer who understands the camera stops needing the vocabulary.

The vocabulary is still worth knowing, because it is what the literature uses and because it names a real feature of a picture. It is just not a set of methods to choose between.

Reading a drawing for its case #

A last practical note: given someone else’s perspective drawing, the case it was constructed in can be read off in a few seconds, and reading it says something about the drawing.

Verticals parallel means a vertical picture plane, so the camera was level and the viewer’s eye height is on the horizon. Verticals converging upward means the camera was tilted up, and the drawing is of something taller than the viewpoint. Horizontals parallel across the front means the object is face-on, and the drawing is depicting a frontal relationship deliberately.

Both horizontal vanishing points comfortably on the sheet means a wide angle of view, which means the drawing is correct only from very close and will read as exaggerated at any normal viewing distance. That last one is the most common tell in student work and the easiest to fix: move the vanishing points further apart, off the paper if necessary, and the drawing settles down.

A fourth case the chapter headings do not have #

One-, two- and three-point are not three methods; they are one projection with the picture plane at three orientations, and the count is the number of axes that are not parallel to it. That framing has a consequence the traditional chapter structure hides: the count can be zero.

A picture surface that is not a plane has no picture-plane orientation to count against. On a cylinder a set of horizontal parallels does not converge to a point at all — it converges to two points, at azimuth ±90°, and the line between them is not straight. On a stereographic surface every straight world line images as an arc of an exact circle, and the two ends of the arc are the two vanishing points, which is a perfectly good projective statement and not a point count.

So the one-two-three taxonomy is a taxonomy of flat pictures specifically. That is worth knowing before it is generalised, because the generalisation people reach for — “a fisheye is four-point perspective” — is not a statement about anything. What a curved surface has instead of a vanishing point count is a set of measured costs: how much it bends a straight line, how far it is from preserving angle, how unevenly it scales area.

And the parameter the count leaves out #

A one-point construction has a free parameter besides its depth: the ratio of its far edge to its near one. Below 1 the meeting point is above the far edge; at 1 there is none; above 1 it is below the near edge.