Taking a frame apart

What the passes hold

A renderer does not produce a single image by accident. It produces one because, by default, it adds every interaction into a single float buffer and hands you the total. Split that addition across separate buffers and you have passes — or, in the older compositing vocabulary, elements. The division is not arbitrary: it follows the terms in the rendering equation itself.

The diffuse pass holds the light that scatters equally in all directions from a surface — the broad, directionless illumination that reads as an object's local colour. It subdivides further: direct diffuse (illumination from lights the renderer can see in one step) and indirect diffuse, also called GI, which is everything that arrived via at least one bounce. Separate them and you can brighten indirect without washing out the highlights that sit in a different pass entirely.

The specular pass holds the directional reflection — the glossy lobe, tight or broad, that carries the visible shape of a light source. Because specular energy is view-dependent, it cannot be multiplied into diffuse the way a surface colour can, so compositors add rather than multiply them together, and that addition is exactly what the renderer was doing internally.

Transmission holds light that has passed through the volume of a surface — glass, liquid, translucent plastic. It is often absent or near-black in shots that contain no refractive objects, which is itself useful to know: an unexpectedly bright transmission pass is a scene geometry error waiting to be diagnosed.

Subsurface scattering is sometimes folded into diffuse, sometimes separated, depending on what the production needs. Skin, wax and marble all carry significant subsurface contributions; having the pass isolated means a colourist can warm or cool it independently of the surface diffuse.

Emission is the simplest pass: every surface that is itself a source of light. In a scene with glowing signage or practical bulbs in frame, emission is the pass you pull down first when the shot reads as overlit.

Reflection and refraction are sometimes named distinctly from specular and transmission to separate environment and inter-object rays from direct-light rays — the taxonomy varies by renderer, so the first task on any new pipeline is to confirm exactly which rays populate which buffer.

It produces one because, by default, it adds every interaction into a single float buffer and hands you the total.

What recombination allows — and what it does not

Adding the passes back together should reproduce the beauty exactly. If it does not, a pass is missing or the blend mode is wrong — and confirming that beauty is the sum of its passes is the check every compositor should run before touching anything. The value of the exercise is not that it always finds a problem; it is that it proves the arithmetic is intact.

Once verified, the passes become handles. Pulling the diffuse down in grade without affecting the chrome in the specular pass — that is a thing compositing can do cleanly. Adjusting the colour of indirect light to feel warmer without affecting direct highlights — also clean. These adjustments are linear operations on linear data, which is why the passes must never be viewed or manipulated in a gamma-encoded context before recombination; the maths only closes when everything is in the same space.

What passes cannot do is separate information that was never recorded separately. If a renderer bakes caustic energy into the diffuse buffer, there is no subsequent operation that cleanly extracts it. If direct and indirect diffuse were summed before output, warming the GI means warming everything in that buffer. The pass structure of a render is a set of promises made at render time, and compositing can only honour what was promised. This is why the conversation about which passes are needed happens before the render farm starts, not after it finishes.