Short answer
Colour correction makes pictures technically consistent: neutral whites, matched exposure, shots that agree with each other. Colour grading is the deliberate look laid on top, the contrast, palette and mood. Both use the same controls, so the difference is intent and order rather than software. How a grade turns out depends on three things the grader has to keep track of: the encoding the image was in when it was adjusted, the transform that turned it into a displayable picture, and the screen it was judged on.
Grading software describes its own work in two layers. Primary correction acts on the whole frame: one maker calls balancing the image the first step, done with controls that affect the entire picture and aimed at a neutral starting point. Secondary correction isolates part of it, either with a qualifier that selects pixels by hue, saturation or luminance, or with a drawn shape that follows an object. Stills work the same way under other names. White balance and exposure in a raw editor are the correction; split toning, colour-range masks and local adjustments are the grade. The order matters because a look built on unbalanced shots has to be rebuilt for every one of them.
A display-referred image, such as a JPEG or a Rec. 709 video file, is a finished picture: its numbers say what a screen should show. A scene-referred image describes the light in front of the lens. Raw files are scene-referred, and so are the logarithmic encodings cinema cameras record. ARRI describes Log C as a scene-based encoding in which the signal rises by a fixed amount for each stop of exposure, which is why ungraded log footage looks flat and desaturated. Its LogC4 specification places an 18% grey at about 28% of the signal, where the sRGB curve would put it near 46%. The same slider therefore does different things in each world: an offset on a log signal behaves like an exposure change, while on a display-referred image it lifts and flattens the blacks.
The C in Log C stands for Cineon, the earlier log encoding modelled on the density of scanned film negative.
The Academy Color Encoding System is a documented way of arranging those stages so that different cameras and different displays meet in the middle. An input transform converts each camera's native encoding into ACES2065-1, a linear, scene-referred encoding with very wide primaries, used for interchange and archive. Grading is done in ACEScct, a logarithmic encoding with a toe that the documentation says was added because colourists expected a lift control to behave as it does on camera log. Look transforms carry technical or creative adjustments. An output transform then renders the scene-referred image for one kind of display and encodes it for that device, so one grade can be delivered to a cinema projector, an SDR monitor and an HDR television through three different output transforms. ACES is a framework productions may adopt, not a requirement.
A lookup table is a sampled function: numbers in, numbers out, with interpolation between the samples. The Academy's Common LUT Format distinguishes a 1D LUT, which maps each channel through its own curve, from a 3D LUT, a cube of output colours indexed by red, green and blue together and read by trilinear or tetrahedral interpolation. Because a 1D LUT never lets one channel influence another, it can change tone, contrast and channel balance but cannot change saturation or move one hue towards another; that needs a 3D LUT or a matrix. In use, LUTs fall into two kinds. A technical LUT converts between encodings, for example a camera's log signal to Rec. 709. A creative LUT is a stored look. Either one assumes a particular input encoding, and feeding it anything else is the commonest way a LUT goes wrong.
A 3D LUT only knows its lattice points. Fine gradients between them are interpolated, which is why a heavy look stored in a small cube can show contouring.
Colour wheels labelled lift, gamma and gain adjust shadows, midtones and highlights in overlapping ranges, with an offset control for the whole image. They are a convention shared across products, and each product implements the maths its own way, so the same wheel positions do not transfer between programs. The ASC Color Decision List exists to fix that for basic corrections. It defines three functions per channel, applied in order: multiply by slope, add offset, raise to power. A tenth number sets saturation about a luma computed with the Rec. 709 weights. Slope pivots around black, offset moves every level by the same amount, and power bends the middle while holding black and white. The table works the formula through for a neutral ramp. A CDL carries primary corrections only; it has no way of describing a qualifier, a window or a curve.
A grade is an adjustment made until a picture looked right on one screen in one room, so it inherits both. Standard-dynamic-range mastering assumes a reference display with the BT.1886 response, close to a 2.4 power curve, at about 100 cd/m² in dim surroundings. A phone at several times that brightness in daylight, a laptop with a different tone response, a wide-gamut panel showing the file without colour management and a television in a vivid picture mode each render the same numbers differently. Shadows open up or block, saturation rises, and the white point drifts warm or cool. None of this is a fault in the file. It is the reason scopes are used to judge the signal, and why a look is worth checking on at least one ordinary screen before it is signed off.
| Input (0–1) | Slope 1.2 | Offset +0.05 | Power 1.2 | Slope 1.2, offset −0.02, power 0.9 |
|---|---|---|---|---|
| 0.00 | 0.000 | 0.050 | 0.000 | 0.000 |
| 0.10 | 0.120 | 0.150 | 0.063 | 0.126 |
| 0.18 | 0.216 | 0.230 | 0.128 | 0.231 |
| 0.50 | 0.600 | 0.550 | 0.435 | 0.612 |
| 0.90 | 1.000 | 0.950 | 0.881 | 1.000 |
| Range below white | 8-bit | 10-bit |
|---|---|---|
| White to −1 stop | 67 | 271 |
| −1 to −2 stops | 51 | 203 |
| −2 to −3 stops | 38 | 151 |
| −3 to −4 stops | 28 | 114 |
| −4 to −5 stops | 22 | 86 |
| −5 to −6 stops | 15 | 63 |
| −6 to −7 stops | 12 | 48 |
| −7 to −8 stops | 9 | 36 |
| Scope | What it plots | Use it to check |
|---|---|---|
| Waveform | Luminance and colour levels across the frame, overlaid | Exposure, black and white levels; overlapping channels read as white where a region is neutral |
| RGB parade | A separate waveform for each channel, side by side | Colour casts (one channel sitting higher) and single-channel clipping |
| Vectorscope | Hue as angle and saturation as distance from the centre | Overall saturation, and whether skin has drifted towards green, yellow or magenta |
| Histogram | How many pixels sit at each level, per channel | Blown highlights and crushed shadows |
| Chromaticity | Colours plotted against a delivery gamut | Whether colours fall outside the target, for example the Rec. 709 triangle |
Why: The source was 8-bit and the grade stretched a range that held only a few code values.
Fix: Grade from raw or a 10-bit or deeper source, keep the move small, or add fine grain to break the steps up.
Why: A saturation or gain push drove one channel to its maximum.
Fix: Watch the RGB parade while adjusting and back off, or reduce saturation only in that hue range.
Why: A global tint or hue rotation moved skin along with the background.
Fix: Keep the primary neutral and apply the look through a secondary that leaves the skin range alone; confirm on the vectorscope.
Why: It was judged on an uncalibrated or over-bright screen, or one showing the file without colour management.
Fix: Calibrate the grading display, work in dim surroundings, trust the scopes, and review on a second ordinary screen.
Why: The LUT expects one camera's log encoding and was given another's, or an image that was already display-referred.
Fix: Convert to the encoding the LUT was built for first, or use a technical LUT for the conversion and the creative one after it.
Each statement is labelled by kind — established fact, a standard’s requirement, observed market data, a convention, or Colourwise’s own interpretation or analysis — with the strength of the evidence behind it.
StandardStrong evidence
The ASC CDL, as reproduced in the Academy's Common LUT Format specification, computes each channel as clamp(in × slope + offset) raised to power, then applies a single saturation value about a luma of 0.2126 R + 0.7152 G + 0.0722 B; slope and saturation must be zero or greater and power greater than zero.
Caveat: Cited from the Academy's specification, which implements version 1.2 of the ASC's equations. The ASC's own document is distributed on request and was not read. Many applications use an unclamped variant.
FactStrong evidence
ACES defines ACES2065-1 as a linear, scene-referred interchange and archive encoding, ACEScct as a quasi-logarithmic working encoding for grading, and an output transform that renders the scene-referred image for a target display and then encodes it for that device.
Source: ACES Documentation: system overview, encodings and output transforms
FactStrong evidence
ARRI's Log C is a scene-based logarithmic encoding in which the signal increases by a fixed amount per stop of exposure, and the LogC4 specification maps a scene-linear value of 0.18 to a signal of 0.2784.
Caveat: One manufacturer's encoding; other camera makers publish different curves.
Source: Log C (ARRI image science); ARRI LogC4 Logarithmic Color Space: Specification
StandardStrong evidence
The Common LUT Format specifies 1D LUTs read by linear interpolation and applied per channel, and 3D LUTs indexed by all three components and read by trilinear or tetrahedral interpolation, and advises against 3D grids larger than 128 points per axis.
ConventionModerate evidence
Grading tools conventionally split work into primary correction of the whole image, using lift, gamma, gain and offset controls over overlapping tonal ranges, and secondary correction of a selection made by hue, saturation or luminance or by a drawn shape.
Caveat: Vocabulary documented from one widely used product; the mathematics behind the wheels is not standardised.
Source: DaVinci Resolve: Color page
StandardStrong evidence
ITU-R BT.1886 defines the reference display response for HD studio mastering, and ITU-R guidance on HDR production takes 100 cd/m² as the nominal peak white of the SDR reference display.
Source: Recommendation ITU-R BT.1886: Reference electro-optical transfer function for flat panel displays used in HDTV studio production; Report ITU-R BT.2408: Guidance for operational practices in HDR television production
Colourwise analysisStrong evidence
An 8-bit sRGB-encoded image has 67 code values for the stop below white but only 15 for the stop between five and six stops down; a 10-bit file has roughly four times as many at every level.
Based on: Counted by Colourwise from the sRGB encoding function in IEC 61966-2-1: the integer codes at successive halvings of linear light, at 8 and 10 bits. See the table.
Source: International Electrotechnical Commission (IEC) webstore and catalogue
Colourwise interpretationLimited evidence
A global hue, tint or temperature move changes skin along with everything else, and viewers notice a small shift in a face sooner than a large one in a wall, so most looks are built from a neutral primary and selective secondaries rather than one global push.
Based on: Colourwise's reading of the primary and secondary workflow documented by grading-software makers, and of the vectorscope's documented use for checking that skin has not shifted towards green, yellow or magenta. Not a controlled study of viewers.
Caveat: A working rule of thumb, not a measured threshold.
Reviewed 1 October 2026. Colourwise summarises its sources in its own words and does not reproduce standards text or proprietary colour data. Spotted an error? Tell us.