The last department that touches the picture
Colour grading is the stage of post-production where the recorded image is turned into the image an audience sees. It is partly technical — matching shots, correcting exposure and white balance, conforming to a delivery standard — and partly authorial, since the grade sets the emotional register of a scene as surely as the score does. It is also the point where every earlier decision about sensors, lenses, lighting and encoding either pays off or has to be rescued.
Understanding grading means understanding the pipeline it sits in. A frame travels from photons hitting a sensor, through an encoding designed to preserve latitude, through creative manipulation, to a transform that maps it onto a specific display. Most confusion about film colour comes from collapsing those stages into one.
What the camera actually records
A digital cinema sensor measures light through a colour filter array and produces raw values that are proportional to scene luminance. Those values are not an image; they are measurements in the camera’s own native colour space, which is defined by the spectral sensitivities of its filters and differs from every other camera on the market.
Two things then happen. The values are converted towards a standard colour space, and they are encoded with a curve. Cinema cameras generally use a logarithmic curve — the various log formats offered by each manufacturer — because human perception of brightness is roughly logarithmic, so a log encoding distributes the available code values efficiently across a wide dynamic range. Log footage looks flat and desaturated on a normal monitor precisely because it has not yet been mapped to a display.
- Raw defers decisions. Recording raw sensor data means white balance, and sometimes even the sensitivity setting, remain adjustable in post with no quality penalty. Recording a baked-in display-referred image does not.
- Log is not a look. Log is a container. Viewing log directly and calling it flat is like reading a compressed archive and complaining that it is not a document.
- White balance is a scene decision. Setting the camera to the dominant illuminant neutralises it. Deliberately setting it elsewhere — cooling a night exterior, warming a domestic interior — is a creative choice that grading can adjust but not fully undo once clipping occurs.
- Cameras disagree. Two bodies from different manufacturers pointed at the same set produce measurably different colour, particularly on saturated reds and on skin under mixed lighting. Multi-camera shoots need a matching pass before any creative work starts.
Scene-referred and display-referred
The single most useful concept in modern grading is the distinction between scene-referred data, which describes light as it existed in front of the camera, and display-referred data, which describes light a screen should emit. Grading systems increasingly work in a wide, scene-referred space and apply the conversion to a display at the very end.
The Academy Colour Encoding System, usually shortened to ACES, formalises this. Footage from any camera passes through an input transform into a common wide-gamut scene-referred space, creative work happens there, and an output transform maps the result to the intended display — a cinema projector, a broadcast monitor, a high dynamic range television. The advantage is that one grade can be re-targeted to several deliverables without regrading from scratch, and that archived masters are not welded to the display technology of their decade.
The colour spaces you will meet
- Rec. 709. The standard for high definition television and the default assumption for most video. A relatively small gamut and a modest peak brightness, but still the most common target.
- DCI-P3. The digital cinema projection standard, wider than Rec. 709 particularly in reds and greens, and also the practical gamut of many modern consumer displays.
- Rec. 2020. The ultra high definition container, wider than any current display can fully reproduce, which means content mastered in it is effectively mastered for future hardware.
- ACEScct and ACEScc. Working spaces with log encodings designed so that grading controls behave the way colourists expect, rather than for storage efficiency.
The tools of the grade
A colourist works through a small vocabulary of operations, layered in a node graph or a stack.
Primary corrections affect the whole frame: exposure and contrast, the classic lift, gamma and gain controls that push shadows, midtones and highlights respectively, and overall saturation and hue. Primaries do the heavy lifting of getting a shot to sit correctly.
Secondary corrections isolate part of the image. A qualifier keys a range of hue, saturation and luminance — commonly skin tones, sky or foliage — and applies a correction only there. A power window is a geometric mask, often tracked to follow a moving subject, used to lift a face out of shadow or darken a distracting corner.
- Work in a sensible order. Balance and match before you stylise. A look applied over unmatched shots multiplies the mismatch rather than hiding it.
- Watch skin above everything. Audiences have no reference for the correct colour of a wall and an extremely precise one for the colour of a face. Many colourists keep a vectorscope with the skin tone line visible at all times.
- Use scopes, not just eyes. Waveform, vectorscope and histogram catch illegal levels and drift that adaptation hides. After twenty minutes on one shot, the eye has normalised to it.
- Keep a lookup table honest. A LUT, a precomputed table that maps input colours to output colours, is convenient but lossy at the edges and often clips values it was not built to handle. Applying a creative LUT early and grading through it bakes in problems that cannot be undone.
High dynamic range and the second master
HDR delivery changed grading from producing one master to producing several. Standard dynamic range video assumes a display of roughly one hundred nits, a nit being one candela per square metre. HDR masters are commonly graded on displays capable of one thousand nits or more, with the perceptual quantiser curve, or PQ, defining an absolute mapping from code value to emitted luminance. The alternative, hybrid log gamma, is relative rather than absolute and was designed to degrade gracefully on ordinary televisions, which suits live broadcast.
The complication is that a grade optimised for a bright, wide-gamut display does not simply scale down. Highlights that read as a subtle glint at a thousand nits become a blown white at one hundred; a shadow that holds detail on an OLED disappears on a phone in daylight. The usual workflow is to grade the HDR master, derive the SDR version through an automated transform, and then hand-adjust it in a trim pass, because the automated result is a starting point rather than an answer.
- Grade the harder target first. Deriving a smaller container from a larger one is generally easier than the reverse.
- Budget the trim. The SDR trim is real creative work, not a render job, and productions that skip it ship a version that most of their audience will actually watch.
- Check on a bad screen. A phone at half brightness in a bright room is a realistic viewing condition for a great deal of content, and it forgives nothing in the shadows.
The room is part of the instrument
A grading suite is a controlled viewing environment, and the control is not decorative. Walls are neutral grey so that reflected light does not tint the display. Surround illumination is set to a defined level and colour temperature, because perceived contrast and colour depend strongly on what surrounds the screen. The reference display is profiled regularly with a probe, and its state is verified rather than assumed.
Screen calibration for reference work targets a specific white point, a specific transfer function and a specific gamut, with tolerances tight enough that two suites should reach the same decision. Consumer displays do none of this, which is why a grade always involves a step of imagining the range of conditions the work will actually be seen in.
Look design, and the complaint about orange and teal
Beyond correction, grading builds a look: a consistent bias applied across a film that carries meaning. Cool, desaturated palettes for institutional coldness. Warm, high-contrast grades for nostalgia. Split toning that pushes shadows one way and highlights another.
The widely mocked orange-and-teal palette is a real tendency with an understandable origin. Skin tones sit in the orange region of the colour wheel, and pushing everything else towards the complementary cyan-blue maximises the separation between actors and background with a single global move. It reads as expensive because it resembles the deliberate colour contrast of well-lit cinematography, and it became a cliche because it is cheap to apply to footage that was not lit that way.
The more interesting looks tend to be built in the lighting and production design and merely finished in the grade. A colourist can strengthen a relationship between colours that exists in the material. Manufacturing one that was never shot generally shows, in the form of haloed edges, crushed shadows and skin that has quietly stopped looking like skin.
Delivery and consistency
The final stage is producing the required deliverables and proving they are correct. That means rendering to specified codecs and containers, verifying levels against broadcast or platform requirements, checking for illegal chroma, and running quality control passes that catch frame-level errors a human watching in real time will miss. Streaming platforms add their own encoding stage after delivery, and heavy compression is unkind to smooth gradients, so a grade with large areas of subtle sky can develop banding that never appeared in the suite. Adding a small amount of grain or dither before encoding remains the standard defence, and it is one of the few cases where deliberately degrading an image makes the delivered version better.
