Visual Cinema Masterclass · Module 4 of 6

Colour, End to End

Following one image from sensor to screen without losing it

Colour goes wrong in the joins. This module follows a single frame through capture, encoding, grading and delivery, and names what can be lost at each handover and why.

Stage one: the sensor does not see colour

A photosite counts light. It has no idea what colour that light was. Colour is produced by covering the sensor with a mosaic of filters so that each site records only part of the spectrum, then reconstructing full colour for every pixel by interpolating from the neighbours. That reconstruction is an educated guess, which is why fine detail and strong colour edges are where sensors reveal their limits.

Everything downstream inherits that guess. This is the first argument for recording raw, which keeps the unreconstructed data and defers the decision, and the first argument against it, which is that the decision has to be made by somebody eventually.

Stage two: fitting a wide world into a narrow file

A sensor can capture a far greater range of brightness than a conventional video file can carry. Logarithmic encoding is the compromise. It distributes the available code values so that more of them describe the shadows and midtones, where the eye is sensitive, and fewer describe the highlights, where it is not.

The consequence is the flat grey image that surprises people the first time they see it. That footage is not broken and is not a look. It is a container holding more range than the screen can show, waiting for a transform that maps it back. Judging it uncorrected is like judging a negative by holding it up to a window.

Two related numbers decide how much survives. Bit depth sets how many distinct steps exist between black and white, and too few produces banding in gradients such as skies. Chroma subsampling reduces colour resolution relative to brightness, on the reasoning that the eye notices detail in brightness far more. That reasoning holds for viewing and breaks down for compositing, which is why keying green screen from heavily subsampled footage is difficult.

Stage three: agreeing what the numbers mean

A number in a file is meaningless until something says what colour it represents. A colour space is that agreement, defining which real colours the extremes correspond to and how the values in between are distributed. Two files with identical numbers and different colour spaces are different pictures.

This is why the same clip can look correct in one application and wrong in another. Nothing has changed the data. Something has changed the assumption about what the data means. Colour managed workflows exist to make that assumption explicit at every handover rather than leaving each tool to guess.

A standardised interchange system takes this further by converting everything into one very wide common space on the way in, doing the work there, and converting out to whatever the delivery requires. The benefit is that cameras from different manufacturers stop being a special case each. The cost is that the setup has to be right, because a mislabelled input is now wrong in a way that is harder to see.

Stage four: the grade, and what a lookup table really is

Grading divides into two jobs that get conflated. Correction makes the material consistent and neutral, matching shots and fixing exposure and balance. Grading then applies an intent on top of that foundation. Doing the second before the first produces a look that falls apart the moment shots are cut together.

A lookup table is a translation list. It says that an input value becomes an output value, and nothing more. Technical tables convert between colour spaces and are close to lossless in intent. Creative tables carry a look. The failure mode is treating a creative table as a starting point rather than an ending, applying it to unbalanced footage and then trying to correct underneath it. Scopes matter here because a table can hide clipping that the eye forgives on a small monitor.

Where this actually goes wrong

Almost every colour problem in practice is a handover problem rather than a grading problem. Footage arrives labelled as one thing and is in fact another. A round trip through a second application strips the tag that said what the numbers meant. A creative table gets baked in early and the shot underneath can no longer be matched to anything.

Two habits prevent most of it. Decide the colour space at the point of ingest and write it down where the next person will see it, rather than relying on each tool to guess correctly. And keep the correction stage separate from the look, so that a shot which needs rebalancing can be rebalanced without unpicking the creative intent sitting on top of it.

Neither habit is technically demanding. Both are organisational, which is why they are usually the thing skipped when a schedule tightens, and why colour is where a rushed pipeline shows first.

Stage five: delivery, where the assumptions get tested

High dynamic range delivery raises the ceiling rather than brightening everything. Its value is in specular detail, the glint on metal and the texture inside a bright window, which conventional delivery simply cannot hold. Grading for it is a different discipline, because the comfortable viewing level and the peak are much further apart.

Whatever the target, the last stage is the one that punishes an unmanaged pipeline. A file that looked correct on the grading monitor and wrong everywhere else usually means an assumption was left implicit somewhere upstream.

Vocabulary

Demosaicing: Reconstructing full colour for every pixel from a sensor where each site recorded only part of the spectrum. An interpolation, not a measurement.
Log encoding: Distributing code values so more of them describe shadows and midtones. Produces flat looking footage that is carrying more range than the screen can show.
Colour space: The agreement that says which real colours a file's numbers represent. Identical numbers in different spaces are different pictures.
Chroma subsampling: Recording colour at lower resolution than brightness. Usually invisible when viewing and a real problem when keying or compositing.
Lookup table: A translation list from input values to output values. Technical ones convert between spaces. Creative ones carry a look and belong at the end, not the start.

Where the detail lives

This module carries the through line. The step by step work is in these guides.

Sources and confidence

Colour standards are defined by published specifications from the relevant standards bodies. This module describes how the stages relate. The linked guides carry the specific values and tested workflows.

Previous: Exposure and Light · Course syllabus · Next: Post Production and Delivery