Short answer
Most grading faults are not creative errors but a broken link in the chain of transforms: one missing, one applied twice, one meant for different footage, or a file read in the wrong signal range or shown with the wrong display response. The remainder are data that was never there, clipped highlights and saturated colours, or too few code values stretched too far. Work out which state the image is in at each stage before touching a colour control.
Every stage of a pipeline expects an image in a known state: this log curve, this gamut, this signal range, scene-referred or display-referred. Faults appear where an assumption is false. So the first diagnostic step is clerical. Write down, for the source, the working space, the viewer and the export, what encoding each believes it has. A table of transforms that looks correct on paper and still gives a wrong picture usually has a second, hidden conversion in it, applied automatically by the software on import or on display.
A sensor or a file has a maximum. When a channel reaches it, detail above that level is gone, and because the three channels clip at different points a bright coloured light drifts in hue as it burns out, usually towards yellow, cyan or white. No grade restores this. The same loss can be created in post-production by an early display-referred step. Dolby's guide warns that highlights clipped in an HDR master cannot be recovered in the SDR version derived from it, which is a statement about every later stage as well.
A colour can be within the range of brightness and still be outside what the delivery space can hold: a deep LED blue, a neon sign, a saturated costume under coloured light. Pushed past the boundary, one channel goes to zero or to maximum and the surface loses its modelling and turns into a flat, posterised patch. Wide-gamut camera spaces hold such colours without complaint, so the fault appears only at the output. The remedy is to compress towards the boundary before reaching it, which is what the gamut compression stage of a modern output transform is for.
Banding is the visible staircase in what should be a smooth gradient, a sky or a wall falling into shadow. It comes from arithmetic: a log recording at eight bits has few code values per stop, and applying display contrast pulls neighbouring values apart until the step between them can be seen. Heavy compression makes it worse by discarding fine variation first. It also comes from coarse lookup tables. Recording and working at ten bits or more prevents most of it; fine grain or dither added before the final encode hides what remains.
Two faults survive a correct grade and appear only on export. The first is signal range. Broadcast video is narrow range by default, with black and white at code values inside the full scale, and a file read the other way is either crushed and clipped or lifted and dulled. The second is display response. A picture graded for the 2.4 response of a reference monitor is brighter in its shadows and mid-tones when a computer shows it at 2.2. Neither is an error in the grade, and neither should be fixed by regrading.
Nothing on this site is a calibrated reference. A web page is shown by a browser that may or may not colour-manage, on a screen of unknown white point, brightness and response, so no still or swatch here can reproduce a graded image.
| Signal level | Light on a 2.4 display | Change at 2.2 | Change at 2.6 |
|---|---|---|---|
| 10% | 0.4% | +58% | -37% |
| 25% | 3.6% | +32% | -24% |
| 50% | 18.9% | +15% | -13% |
| 75% | 50.1% | +6% | -6% |
| 90% | 77.7% | +2% | -2% |
Why: A scene-referred log image is being shown or exported with no output transform, or the transform was bypassed.
Fix: Add the correct output transform for the camera's encoding and the delivery standard; do not try to build contrast by hand.
Why: A transform has been applied twice: once automatically on import or in the viewer, and again as a lookup table or node.
Fix: Disable one. Check the software's input colour management settings as well as the visible grade.
Why: The transform is for a different log curve, gamut or exposure-index variant from the one the footage was recorded in.
Fix: Read the clip metadata and match curve and gamut exactly; do not correct the error with colour wheels.
Why: One or two channels clipped at the sensor or at an early display-referred step.
Fix: If it is in the recording it cannot be restored, only disguised by desaturating the clipped area. If it is created in the pipeline, move the offending step later.
Why: The colour is outside the delivery gamut and has been clipped at its boundary.
Fix: Apply gamut compression or reduce saturation selectively before the output transform, keeping a gradient across the surface.
Why: Too few code values for the stretch applied: 8-bit or heavily compressed log, or a coarse lookup table.
Fix: Work at 10 bits or more from the original files, use a denser table, and add fine grain before the final encode.
Why: Signal range mismatch: a narrow-range file read as full range, or the reverse.
Fix: Set the range flag at export and the range interpretation in the player to agree. Do not regrade to compensate.
Why: The grade was made for a 2.4 reference display in a dim room and is being shown at about 2.2, or the player reinterprets the file's transfer tag.
Fix: Confirm the file's colour tags, compare on the reference monitor, and decide deliberately whether a separate web deliverable is needed.
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.
FactStrong evidence
Dolby's grading guide states that highlights clipped in the HDR master cannot be recovered in the Rec. 709 SDR target derived from it.
Source: Dolby Vision Color Grading Best Practices Guide, version 4.0
StandardStrong evidence
ITU-R BT.2100-3 describes narrow-range signal representation as the one in widespread use and the default, and says the full-range representation should not be used for programme exchange unless all parties agree.
Colourwise analysisStrong evidence
A 10-bit narrow-range file read as full range shows black at about 6 per cent and white at about 92 per cent of the scale; a full-range file read as narrow puts black about 7 per cent below zero and white nearly 10 per cent above full scale, both of which are clipped.
Based on: Calculated by Colourwise from the nominal 10-bit narrow-range limits of 64 and 940 against the full scale of 0 to 1,023; see the table on the broadcast-legal page.
Source: ITU-R Recommendations BT.709, BT.2020 and BT.2100; EBU R 103: Video Signal Tolerance in Digital Television Systems, version 3.0
Colourwise analysisStrong evidence
A display with a 2.2 power response emits about 15 per cent more light at a half-scale signal, and about 58 per cent more at a 10 per cent signal, than the 2.4 response of the BT.1886 reference display.
Based on: Calculated by Colourwise as the ratio of the signal raised to 2.2 and to 2.4, ignoring black-level terms; see the table on this page.
Caveat: Pure power laws with zero black. Real displays, and the sRGB curve many computers approximate, differ most near black.
StandardStrong evidence
The EBU cautions that gamut legalisers may create artefacts more disturbing than the errors they correct, and advises against legalising before all other signal processing is complete.
Source: EBU R 103: Video Signal Tolerance in Digital Television Systems, version 3.0
Reviewed 6 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.