All posts
By Pravit Gandhi··8 min read

Why does my grade look off?

Why does my grade look off? Six causes share that symptom. Isolate them in order, starting with the file and ending with your own eyes.

Off is not a diagnosis. At least six separate things produce the feeling that a grade is wrong without being wrong anywhere you can point at, and they are not equally easy to check: the file you are judging, the display, the room around the display, the tool you built contrast with, what happened to the faces, and how long you have been sitting there. Work down that list in order. The first two are cheap and either confirm or clear themselves in a couple of minutes. The last one cannot be ruled out at all while you are still inside it, which is why it belongs at the bottom rather than the top.

Start with the file, not the grade

Before you change a single wheel, find out whether the picture you are unhappy with is the picture you made.

Export a short section and open it in two players: one that reads color tags and one that does not. If they disagree, nothing you do in the grading tool will fix it, because the difference is being introduced after your last node. Apple's Technical Note TN2227 describes the mechanism, where a QuickTime file carries an nclc tag declaring primaries, transfer function and matrix, and ColorSync builds a transform from that tag to your display. The failure case is in the same document: media without an nclc tag is color managed "as if it were created in the SMPTE-C color space," which is not what anyone grades in.

This goes first precisely because it feels like the least likely explanation. If the file looks correct in the tag-ignoring player, your grade is fine and you have an export problem, covered in why S-Log3 footage looks washed out.

Then check what the display is doing to it

Assume the file is honest. The next link is the only thing standing between the numbers and your opinion.

Blackmagic is unusually blunt about how much rests on this. Describing the default DaVinci YRGB color science, the Resolve 20 manual says the application "has no information about how the source media used in the Timeline is supposed to look; you can only judge color accuracy via the calibrated broadcast display you're outputting to. Essentially, you are the color management." Without a calibrated display there is no reference, and without a reference there is no such thing as a grade being off, only a grade you happen not to like today.

There is a published target to calibrate toward. ITU-R BT.1886 specifies the reference electro-optical transfer function for flat panel displays used in HDTV studio production, and exists because, in its own words, "CRT reference displays are no longer available." Manufacturers publish their own version of the same instruction: Fujifilm's GFX ETERNA technical guide lists the monitor settings it expects when you apply its conversion LUTs, which are colour space Rec.709, colour temperature D65, gamma 2.2.

Two mundane faults are worth eliminating here too. Check the monitoring bit depth, since Resolve's manual notes that monitoring in 10-bit is "preferable to avoid the appearance of banding that may not in fact be in the image data being processed by DaVinci Resolve." And check your computer display's colour profile, because a viewer transform and an operating system profile can both sit in the path at once. Banding has its own causes, in why banding appears after grading and export.

Then look at the room

If the file and the display both check out, the next variable is not on the screen. ITU-R Report BT.2390-10 states the reference condition directly: the viewing environment "would ideally be dimly lit, with the area surrounding the monitor being a neutral grey (6 500 degree Kelvin) at a brightness of 5 cd/m2." The report also explains why anyone bothers specifying that. Rendering intent, the system's deliberate non-linearity between camera and display, "is needed to compensate for the psychovisual effects of watching an emissive screen in a dark or dim environment, which affects the adaptation state (and hence the sensitivity) of the eye."

So a surround change is a picture change without the picture changing. BT.2390 takes that seriously enough to publish a correction for it in the HLG case, giving system gamma for surrounds brighter than the 5 cd/m2 reference as a function of ambient luminance. Those particular numbers belong to HLG and should not be carried into an SDR Rec.709 job, but the reason behind them carries over completely, and the report notes that many programmes are made in environments that "differ considerably from the reference viewing environment."

This is also the usual route into over-saturation. Working against a dim or badly lit display, a grade that felt right in the room reads heavy everywhere else, and the instinct is to add saturation until the picture has some life in it. The adaptation mechanism above is documented; the habit of over-saturating to compensate is a pattern rather than a measured effect, so test it rather than take it on my say-so. Grade a shot, leave it overnight, open it in a different room in the morning.

Then check how you built the contrast

Ruling out the file, the display and the room leaves the grade, and the first thing to look at is not how much contrast you added but which control added it.

Applying a curve to each colour channel independently is not a neutral operation. BT.2390 states it plainly for the display case: "Simply applying a gamma curve to each component separately as is done for SDR television distorts the colour; in particular, it distorts saturation but also to a lesser extent the hue." The report works the arithmetic. Take a pixel whose normalized red, green and blue are 0.25, 0.75 and 0.25, and apply a gamma of 2 by squaring each component. You get 0.0625, 0.5625 and 0.0625. The pixel got darker, and the ratio of green to red and blue went from 3:1 to 9:1, so the green "would have appeared as a discernibly different shade of green."

That is what an aggressive per-channel S-curve does everywhere at once, and it is why the same apparent contrast reads clean from a defined transform and dirty from a curve stacked on an already normalized picture. Which method you used also changes the starting point, since a colour space transform, an ACES pipeline and a manufacturer LUT each answer the display-mapping question differently, as worked through in why CST, ACES and LUTs look different.

Then look at the faces

Skin is where the previous section becomes visible, because faces move through a range of brightness inside a single shot and the eye has strong opinions about all of them.

Pick a face that is lit unevenly, or a shot where somebody walks from shade into sun. Sample the same person in the bright part and the dark part and compare the hue, not the exposure. If the hue drifts as the face brightens, particularly toward orange in the highlights, that is the distortion described above rather than a white balance error, and reducing overall saturation afterwards will not undo it, because the shift happened per channel on the way up.

The other face-related version of off is not about your grade at all: two cameras that were never going to match, side by side in the same scene. That has a different fix, in why your cameras don't match.

Then consider that you have been at it too long

The last item is the one nobody wants, and the reason "demoralized" turns up in the thread titles.

I am not going to dress this up with a study, because I do not have one to cite. What is documented is the underlying capability: BT.2390 describes the eye's adaptation state changing with what surrounds it, the same faculty that lets you stop noticing a colour cast that has been in front of you for two hours. Extend that from a room to a session and the consequence follows without needing a number attached. A grade you have stared at since morning is being judged by an observer who has adapted to it.

The habits that work against this are unremarkable and they do work. Compare against an untouched reference clip rather than your memory of ten minutes ago. Stop before you think you need to, and reopen the timeline cold. If it looks fine the next morning, the problem was never in the timeline.

When none of the six explains it

Occasionally the grade is technically clean and still does not land, and the honest answer is that the reference is doing something your footage cannot. A film still carries halation, grain structure and lens character that no tone curve transfers, so a match on tone and hue relationships still reads differently. That gap is separated out in recreating a movie look from a still.

Leumos AI, our browser-based grading studio, removes one item from this list rather than all six: uploads get an ACES-based input transform for their format, so every shot starts from the same normalized point. It is in closed beta with a waitlist at leumos.ai.

Frequently asked questions

Why does my grade look fine in my editor and wrong everywhere else?

That is the export and tagging case, not a grading error. Apple's TN2227 describes how a QuickTime file's nclc tag drives the transform applied at playback, and states that untagged media gets treated as SMPTE-C. Play the export in a player that ignores tags. If it looks right there, the grade is intact and what needs to change is the export.

How dark should the room be when I grade?

ITU-R Report BT.2390-10 gives the reference condition as a dimly lit room with the area surrounding the monitor being a neutral grey at 6500 K and 5 cd/m2. The point is less the exact figure than that the surround is a controlled part of the system rather than decoration.

Why do skin tones go orange when I add contrast?

Because a curve applied to each colour channel separately does not preserve the ratios between them. ITU-R BT.2390 states that per-component gamma "distorts saturation but also to a lesser extent the hue," and works an example where squaring a pixel's components shifts the green-to-red-and-blue ratio from 3:1 to 9:1. Faces show it first because they sit where the curve is steepest.

Is it worth calibrating a consumer monitor?

It gives you a reference you did not have, which is what makes off a solvable question rather than a mood. Blackmagic's own framing is that with display-referred colour science you are the colour management, working in conjunction with a display calibrated to ensure accuracy.

Sources