← Blog

Right Brain · Reading

Why Flat Pictures Look Three-Dimensional

Hold a photograph. It is a flat sheet with a thin film of pigment on it — no depth anywhere in the object. And yet you do not see a patterned rectangle. You see a room receding, a face standing forward of a wall, a road going somewhere. The interesting question is not why a picture fools you. It is that the same machinery is running when you look at the actual world, and it has never had access to depth either. Take in the whole. Before the parts.

By ComunAI · 10 August 2026 · ~6 min read

The problem your eyes cannot solve

The back of your eye is a surface. Light lands on it in two dimensions, and that is the entire input. A point three metres away and a point thirty metres away, if they line up, arrive at the same place on that surface with nothing to distinguish them. The third dimension is not compressed in the signal — it is absent from it.

So depth is not something you receive. It is something the visual system reconstructs, from a collection of partial hints none of which is decisive on its own. A photograph works because it reproduces most of those hints faithfully. It does not trick you into seeing depth; it supplies the same evidence a real scene supplies, and your system draws the same conclusion it always draws.

The cues a flat picture can carry

These are the pictorial cues — the ones that survive being printed. They were catalogued by Renaissance painters long before anyone measured them, which is itself a good sign that they are real regularities rather than theory.

• Occlusion. If one shape interrupts the outline of another, it is nearer. This is the bluntest cue and one of the most reliable — but it only ever tells you an ordering, never a distance.
• Relative and familiar size. Two objects you know to be similar, one smaller on the page, get read as one being further away.
• Linear perspective and texture gradients. Parallel edges converge; a surface's texture gets finer and denser with distance. James Gibson made texture gradients central to his account of how surfaces specify layout.¹ Strong
• Shading. Vilayanur Ramachandran showed that ambiguous shaded blobs are resolved into bumps or dents according to an assumption that light comes from above — and that flipping the image flips the interpretation.² Strong The system carries a prior about the sun and applies it without consulting you.
• Aerial perspective. Distant things are hazier and shifted toward the colour of the atmosphere between you and them.

Stack enough of these in one image and the reconstruction is not optional. You cannot decide to see the photograph as flat any more than you could decide to see the checker-shadow squares as the same grey — the same stubbornness we went through in Why Optical Illusions Fool Everyone.

The two cues a picture cannot fake

Some depth information depends on you having two eyes, or on you moving. A flat print cannot supply either, which is exactly why it never quite convinces.

Binocular disparity. Your two eyes see slightly different images, and the size of the mismatch for a given feature is a distance signal. Charles Wheatstone demonstrated this decisively in 1838 by building a stereoscope: present each eye with a drawing from the appropriate viewpoint, and observers experience solid depth from two flat pictures.³ Strong

Béla Julesz then removed every other possibility. His random-dot stereograms contain no recognisable shapes, no outlines, no shading — each eye sees only a field of noise. Yet when the two fields differ by a shifted region, a shape floats out in depth.⁴ Strong Depth is extracted before anything is recognised, from raw disparity alone. That result settled a long argument about whether you need to identify an object before you can place it.

Motion parallax. Move your head and nearer things sweep across your view faster than far ones. Brian Rogers and Maureen Graham showed this works as a depth cue in its own right — observers viewing with one eye got a compelling impression of a three-dimensional surface from motion alone.⁵ Strong

Which cue wins, and when

Since the cues can disagree, the system must be weighting them. James Cutting and Peter Vishton's analysis is the useful one here: rather than ranking the cues absolutely, they rank them by distance.⁶ Moderate Within arm's reach, disparity is powerful. At conversational distance it is already weakening. Across a valley it is worthless, and the pictorial cues — occlusion, relative size, aerial perspective — carry the whole load.

Which explains something you can check yourself. A photograph of a landscape can be genuinely convincing; a photograph of a mug on your desk always reads as a picture. In the far field the print supplies everything your eyes would have had. Up close it is missing the cue that mattered most.

When the assumptions are wrong

The clearest evidence that depth is inferred rather than measured is what happens when the inference is deliberately mistargeted. The Ames room, built by Adelbert Ames Jr. and documented by William Ittelson, is a distorted trapezoidal room whose far wall is slanted away — viewed through a fixed peephole, it projects to your eye exactly as a normal rectangular room would.⁷ Strong

Given that projection, your system chooses the enormously more probable interpretation: an ordinary room. Everything else then has to bend around that choice, so two people of the same height standing in opposite corners appear wildly different in size. You are not seeing an error. You are watching a very good inference machine commit to the likeliest reading of a scene that was engineered to be a lie.

The Ponzo illusion is the same principle in miniature: two identical lines across converging rails, and the upper one — which perspective implies is further away, and therefore larger to project that size — looks longer.

What this means for Right Brain

Right Brain is named for a metaphor, not a hemisphere — the open, whole-first way of looking, as against the narrow, part-by-part one. Depth is possibly the cleanest example in all of perception of why whole-first is not a slogan. No single cue in a scene contains the distance. Occlusion gives an order, size gives a ratio, shading gives a curvature — and the layout only exists once they are pooled. Crop the scene down to a patch and the depth genuinely disappears, because it was never in the patch.

The perception games in the app are that idea shrunk to a few quiet minutes: no score, no verdict about your abilities, just your own inference machinery caught briefly in the act. And the standing limit applies as always: there is no good evidence that playing perception games improves your everyday seeing. Broad transfer is the weak link in this whole field, and we covered it properly in Do Brain-Training Games Actually Work? Contested