← Home
GLOSSARY

Published · Updated

Color Grading

Color grading is the post-production process of deliberately altering a video's color and tonality to give it a consistent, intentional look. It works on three things: tone — how bright each range of the image sits, from the black point through the midtones to the highlights; saturation — how far colors sit from neutral grey; and color balance — which hue the shadows, midtones and highlights each lean toward. It is conventionally distinguished from color correction, the pass that comes first and is corrective: neutralizing casts, fixing exposure, and matching shots to one another so white reads as white and skin reads as skin. Correction makes footage accurate; grading makes it look like something. In practice both happen in the same tools and often in the same pass, so the boundary is one of intent rather than technique.

Correction First, Then the Grade

The two words are used interchangeably, and the distinction is still worth keeping, because it encodes an order of operations rather than a difference in tooling.

Correction has a reference. A neutral card should read neutral; skin should land near the skin-tone line a vectorscope draws at roughly 123°; the black point should sit at the bottom of the legal range and not below it. Under Rec. 709 in 8-bit studio swing that range is code 16 (black) to 235 (white) — 219 code values across 0 to 100 IRE, so one IRE is about 2.19 code values; in 10-bit it is 64 to 940. Picture information pushed below 16 is not "very dark", it is gone. Most of correction is unglamorous: killing a cast, matching a second camera to the first, recovering a window that clipped. Grading has no reference, only an intention.

The order matters because errors compound underneath a look. Grade a shot that is still 400K too warm and every later adjustment is judged through that warmth — you end up correcting it with controls that were never meant to, and the compensation shows on any other display. The word itself is a fossil of this: in the photochemical era a lab technician set printer lights per shot on a Hazeltine analyzer, and the job was called color timing or grading long before anyone had a color wheel.

The Controls, and What Each One Actually Moves

Every grading interface — wheels, curves, sliders, a text prompt — resolves to a small number of operations on pixel values. It is worth knowing which one you are actually reaching for.

Exposure, lift, gamma, gain

These are all the tone curve, moved at different points. Lift moves the black end, gain scales the white end, gamma bends the middle without moving either end, and exposure moves the whole thing. The industry writes this down as the ASC CDL: out = (in × slope + offset)^power, applied per channel, with nominal values of slope 1.0, offset 0.0 and power 1.0. Slope is gain, offset is lift, power is gamma. Version 1.2 added a tenth parameter, saturation, applied across the three channels together. Ten numbers is the whole primary grade — which is why a CDL travels in an EDL, an ALE or a FLEx file without carrying any picture.

Contrast, and its pivot

Contrast steepens the curve around a pivot, conventionally middle grey — the 18% reflectance card, which lands near the centre of a gamma-encoded signal. Raising contrast pushes everything below the pivot down and everything above it up simultaneously, so it crushes shadows and blows highlights in one move. This is why "add contrast" is usually the wrong answer to "it looks flat": the fix that actually reads is an S-curve, which steepens the midtones and leaves the two ends alone.

Saturation

Saturation scales each pixel's distance from the neutral axis. Under Rec. 709 that axis is the luma line Y' = 0.2126R + 0.7152G + 0.0722B — green carries roughly 71% of perceived brightness and blue only 7%. That asymmetry has a visible consequence: desaturate a saturated green and it goes light grey; desaturate a saturated blue and it goes nearly black. It is also why a black-and-white conversion never looks like the color image with the color removed.

Temperature and tint

White balance needs two axes because light sources sit in a plane, not on a line. Temperature is the blue-to-amber axis and tracks the blackbody curve: tungsten around 3200K, midday daylight around 5600K, open shade above 7000K. Tint is the green-to-magenta axis perpendicular to it, and it exists because real lights are not blackbodies — fluorescent tubes and cheap LED panels sit off the curve on the green side. That is the single most common defect in self-recorded indoor video, and no amount of temperature adjustment fixes it, because it is not a temperature problem.

What a LUT Is, and What It Is Not

A LUT is a lookup table: a precomputed mapping from input color to output color. A 1D LUT is three independent curves, one per channel — it can move tone and per-channel balance, but it cannot change how colors relate to each other, so it can never make one hue rotate while another stays put. A 3D LUT is a lattice of output RGB values indexed by input RGB, and it can do anything a fixed mapping can.

The lattice is coarse and the gaps are interpolated. Common cube sizes are 17³ (4,913 entries), 33³ (35,937) and 65³ (274,625); most implementations interpolate trilinearly between the eight surrounding lattice points, though tetrahedral interpolation is preferred where neutrals matter, because it does not drift grey off-axis. The de facto interchange file is the plain-text .cube: a LUT_3D_SIZE header and then one RGB triplet per line, in order.

The thing to understand is that a LUT is not a grade. It is a fixed function with no knowledge of your shot. Two clips exposed a stop apart get exactly the same mapping, so a LUT only looks right on footage already matched to the input it was built for. That is the source of the most common failure in amateur grading: dropping a creative look LUT onto raw log footage. Log needs a conversion LUT first — a technical transform to Rec. 709 — and the creative one goes after it. Stacked in the wrong order, or stacked at full strength on top of each other, the second LUT clamps whatever the first pushed outside its input domain, and that detail does not come back.

Why "Cinematic" Means Desaturate and Lift the Shadows

"Make it cinematic" is the most common grading request there is, and it is defined entirely by what it is running away from. The reference point is not a film — it is the footage the person is looking at. Phone cameras and default Rec. 709 picture profiles are tuned to look impressive on a small, bright screen in daylight: saturation up, contrast hard, blacks crushed to zero. That is the "video" look.

Film differs from it in two ways that are physical rather than stylistic:

  • A print cannot reach black. The print stock's base density plus fog still passes light, so the darkest area of a projected image is a dark grey. Digital sensors clip to zero. A lifted black point is not a filter — it is the literal signature of a photochemical print.
  • Film desaturates as it approaches its limit. Emulsion rolls off into the highlights and loses saturation on the way; digital clips hard, channel by channel, which produces the magenta and cyan fringes you see in a blown sky.

So the recipe everyone converges on is a direct imitation: pull saturation down by roughly 10 to 25%, lift the black point so shadows sit around 3 to 6 IRE rather than 0 — code 22 to 30 in 8-bit studio swing — put a gentle S-curve through the midtones, and usually push the shadows toward blue while leaving skin warm. That last part is teal-and-orange, dominant in Hollywood from about 2010 onward, and it works because skin is inherently orange and its complement separates a subject from a background at zero cost in lighting.

The honest version of the advice: cinematic is a restraint, not an addition. Films look expensive because of lighting, lens choice, blocking and production design. The grade is the last ten percent of that, and it cannot add a key light that was never there. What a grade genuinely does for consumer footage is take away the aggressive defaults — which is why the answer to "make it look like a movie" is almost always less of something.

A Scenario Where the Order Bites

You record a 20-minute talking head at your desk. Overhead LED panel, a window off to one side, a plain wall behind you. It looks cheap on playback, and the instinct is to reach straight for a cinematic look.

There are two problems in that frame and only one of them is a look. The wall reads faintly green, because the LED sits off the blackbody curve. And the two sources disagree — the window is around 5600K, the panel around 4000K — so one side of your face is cooler than the other. Apply the look now and you have a green-cast, mixed-temperature image with a cinematic grade on top of it, and every adjustment afterward is you trying to talk yourself out of the green.

The order that works: tint toward magenta until the wall is genuinely neutral, temperature until the face reads consistent, exposure up if the panel underlit you — then the look. And check it cold. Your visual system adapts out a color cast in well under a minute of staring at it, which is why colorists work against a neutral grey surround and keep toggling the grade off. The version you approve after twenty minutes of tweaking is the version you have gone partially blind to.

Common Misconceptions

"Grading and correction are the same thing."

Same tools, often the same pass, but not the same job — and the difference is an order constraint rather than a vocabulary preference. Correction has a target you can be measurably wrong about. Grading does not. Doing them in the wrong order buries a measurable error underneath an unmeasurable one.

"A LUT is a grade."

A LUT is a fixed function of pixel value. It knows nothing about your exposure, your white balance or your subject, so it cannot compensate for a shot being a stop under. A pack of film-emulation LUTs is not a grade; it is the last step of one, and it only lands on footage that already matches the input it expects.

"Cinematic means turning the saturation up."

It almost always means turning it down. Saturated, crushed and contrasty is the default look of consumer cameras and social apps — precisely the thing "cinematic" is a reaction against. If a grade makes footage look more like a phone shot, it is going the wrong way.

"Grade first, then cut."

Shot matching is only meaningful against the shots that end up adjacent, and you do not know which those are until the cut is locked. Grade first and you grade material you throw away — and every later change to the edit puts two differently-graded shots against each other.

"Log footage is broken and needs fixing."

Log is an encoding, not a look. It spends its code values on dynamic range — roughly 12 to 15 stops — rather than on a viewable contrast curve, which is why it arrives flat and desaturated. It is not underexposed and not damaged; it is unfinished by design.

"If it looks right on my screen, it is right."

An uncalibrated laptop panel and an OLED phone disagree about black level, gamma and primaries by enough to change a grading decision. That is the real argument for scopes: a waveform does not adapt, get tired, or run at a different brightness than the viewer's.

How This Works in Valmera

Valmera is an agentic video editor, so grading is described rather than dialed. The agent writes two fields onto the edit decision list, renders a preview, and looks at real frames from it to judge the result — no pixels are touched until render, and every grade is a version you can reverse by asking.

  • Six presets: vibrant, warm, cool, bw, vintage and cinematic — or none to clear.
  • Seven continuous axes on top, applied after the preset so the two compose ("cinematic but warmer" is one preset plus one axis): exposure −1 to +1, contrast 0.5 to 1.6 (1.0 neutral), saturation 0 to 2 (1.0 neutral), temperature −1 cool to +1 warm, tint −1 green to +1 magenta, shadows −1 to +1 (positive lifts the darks), highlights −1 to +1 (negative recovers blown areas). An axis's neutral value clears it.
  • Shadows and highlights are a tonal curve, not a brightness offset. It pins its midpoint at 0.5 and moves the quarter- and three-quarter points, so lifting the shadows raises the black point — at maximum to 0.12 of full scale — while leaving middle grey where it was. That is the difference between the faded-print look above and simply brightening the picture.
  • Captions and graphics are never graded: the grade lands after the cuts assemble and before text burns in. Grain and vignette are separate stylize effects, each able to run over a time window.

The honest limits. No LUT import (no .cube) and no log-to-Rec. 709 transform. No scopes, no color wheels, no secondaries — no HSL qualifier, no power windows, no per-object isolation. And the grade is a single primary applied to the whole program: there is no per-shot grade, so there is no shot matching between clips filmed under different light. For scope-accurate delivery or a multi-camera match, a dedicated colorist application is the right tool. The color grading tool page and the effects documentation carry the working detail.

Related Terms

Other entries in the Valmera video editing glossary that touch this one:

  • Shot detection — Finding the cuts inside a source — the boundaries a per-shot grade would need in order to exist.
  • Proxy editing — Judging a grade on a light stand-in, then applying the decision to the original at full quality.
  • Edit decision list — Where a grade lives when it is a decision rather than baked pixels — which is what makes it reversible.
  • B-roll — The second camera or second source that a grade has to be matched to, and usually was not lit the same way.
  • Burned-in captions vs subtitles — Why the order of grade and caption burn matters: text should not inherit the footage's color treatment.
  • LUFS and loudness normalization — The audio equivalent of grading to a delivery standard — a measured target rather than a taste.
  • Aspect ratio — The other whole-program finishing decision, and the one people confuse with a look.
  • Jump cut — A cut that exposes mismatch — including color mismatch, which is what shot matching exists to hide.

Frequently Asked Questions

Color grading is the post-production process of deliberately altering a video's color and tonality to give it a consistent, intentional look. It works on three things: tone — how bright each range of the image sits, from the black point through the midtones to the highlights; saturation — how far colors sit from neutral grey; and color balance — which hue the shadows, midtones and highlights each lean toward. It is conventionally distinguished from color correction, the pass that comes first and is corrective: neutralizing casts, fixing exposure, and matching shots to one another so white reads as white and skin reads as skin. Correction makes footage accurate; grading makes it look like something.
Intent, and order. Color correction is corrective: it neutralizes color casts, sets a sane exposure and black point, and matches shots to each other so a two-camera interview does not shift color every time it cuts. Its target is a reference — white should be white. Color grading is creative: once the image is accurate, you push it somewhere on purpose, and the target is a feeling. The distinction is not enforced by software — both are done with the same tools, often in one pass — but the order is a real constraint. If you apply a look to a shot that is still 400 kelvin too warm, the error is now underneath every decision you make afterward, and you will spend the rest of the grade fighting it.
A LUT (lookup table) is a precomputed mapping from input color to output color. A 1D LUT is three independent curves, one per channel — it can move tone and per-channel balance but cannot change how colors relate to one another. A 3D LUT is a lattice of output RGB values indexed by input RGB; a 33x33x33 cube holds 35,937 entries and values between lattice points are interpolated, usually trilinearly. A LUT is not a grade. It is a fixed function with no knowledge of your shot: two clips exposed a stop apart get exactly the same mapping, so a LUT can only look right on footage already matched to the input it expects. LUTs come in two categories — technical conversion LUTs, which transform log footage to a viewable Rec. 709 signal and are effectively mandatory, and creative look LUTs, which are an aesthetic and belong after the conversion.
Because "cinematic" is defined by contrast with the thing people are trying to escape: consumer video, which is tuned to look punchy on a small bright screen — high saturation, hard contrast, blacks crushed to zero. Film is the opposite in two measurable ways. A film print physically cannot reach zero: the print stock's base density still passes light, so the darkest area of a projected image is a dark grey, not black. And film desaturates and rolls off as it approaches its highlight limit rather than clipping. So the standard recipe — pull saturation down roughly 10 to 25 percent, lift the black point so shadows sit around 3 to 6 IRE instead of 0, put a gentle S-curve through the midtones — is a direct imitation of a photochemical print. Past about 10 IRE the lift stops reading as film and starts reading as fog or a broken black level.
Exposure moves the whole tone curve up or down. Contrast steepens or flattens it around a pivot, usually middle grey — a contrast control without a pivot just moves where the image clips. Saturation scales each pixel's distance from the neutral axis, which under Rec. 709 is the luma line Y' = 0.2126R + 0.7152G + 0.0722B; saturation 0 collapses everything onto it, which is why a desaturated green looks bright grey and a desaturated blue looks nearly black. Temperature is the blue-to-amber axis, the one that corresponds to the color of the light source: tungsten is about 3200K, midday daylight about 5600K, open shade above 7000K. Tint is the green-to-magenta axis perpendicular to it, and it exists because real lights sit off the blackbody curve — fluorescent and cheap LED panels push green, which is the single most common cast in indoor video.
After. Grading is a finishing operation for a structural reason, not a stylistic one: shot matching is only meaningful against the shots that actually end up next to each other in the program, and you do not know which those are until the cut is locked. Grading first also means grading footage you are about to throw away, and it means every later change to the edit puts two differently-graded shots against each other. The one thing worth doing early is correction on obviously broken material, so you can judge the edit on an image that is not lying to you.
Because it is an intermediate encoding, not a look. A log profile — S-Log, C-Log, V-Log and their relatives — spends its available code values on dynamic range rather than on a viewable contrast curve, preserving highlight and shadow detail across roughly 12 to 15 stops that a standard Rec. 709 profile would clip away. The flat, desaturated appearance is the cost of that: the image is not meant to be watched as recorded, and it requires a transform to Rec. 709 before any creative grading begins. Shooting log is a decision to do more work in post in exchange for more latitude there.
It can apply grading operations from a described intention, and it can judge the result by looking at the rendered frames. Valmera's agent does exactly that: it writes a grade preset and a set of continuous axes onto the edit decision list, renders a preview, and looks at real frames from it. What AI does not replace is a calibrated display, scopes, or shot-by-shot secondary work on a multi-camera shoot. Valmera grades the whole program as a single primary grade and has no LUT import, no scopes and no secondaries — for scope-accurate broadcast delivery or per-shot matching, a dedicated colorist application is the correct tool.

Grade a Video by Describing It

Upload your footage, say what it should feel like, and the agent applies the grade and shows you the frames. 50 free credits, no card.

Start free →
See pricing →

Related Articles

Color Grade a Video
The tool page: six presets, continuous axes and one-request looks, applied by describing them.
Effects Documentation
Grades, stylize effects, looks and censoring — what each one does and what it composes with.
Agentic Video Editor
The category page: what it means for the agent, rather than you, to be the one holding the tool.
MCP Tool Reference
Every tool the connector exposes, including the grading tools and the ranges they accept.