When Should You Color Grade in Log vs. Linear?

Most of my articles are about doing some things in a log space and other things in a linear space. But one thing that I have never covered is the difference between log and linear color spaces. When should we use log versus linear space? What is the difference? And what tools should we be using in both spaces?

I’m going to answer those questions today. My goal is to give you a hands-on, applied, and practical guide to log vs. linear without a bunch of complicated color science or math. I just want to look at how these two spaces work, how they behave, and how we can get the most out of each using DaVinci Resolve.

How I adjust contrast in log

Let’s start by looking at log space. For anyone who’s unfamiliar, log is the default grading space used in scene-referred, color managed workflows, which is how I always work. For this article, I’ll be working in the DaVinci Wide Gamut Intermediate working color space, which is indeed a log space. 

When I think about what I like to do in a log space that I wouldn’t want to do in a linear space, the first thing that comes to mind is contrast adjustments. Let’s take a look at an example so you can see what I mean.

Take a look at the shot above. The exposure feels fine, but I want to shape the low end of the image a little bit. To do this, we will first add a serial node. 

Next, I’m going to confine things to the area below middle gray. To do that, I’ll go to the Gallery, go to my utilities, and add my handy DaVinci Wide Gamut gray patch to the new serial node.

This is simply my exposure chart DCTL (DaVinici Color Transform Language) and you can get a free download of it here. I’ve set its Total Steps value to 1 and I’ve turned off Show Ramp. Dragging this DCTL onto my new node gives me a pure patch of DaVinci Wide Gamut gray. 

The next step is to add another serial node. Then, go to Custom Curves and gang the channels together by clicking the chain icon. 

After that, use the eyedropper tool and click on the DWG gray patch in the timeline. This should give you an anchor point right on that tooth on the lower end of the Custom Curves. 

Now that we’ve done that, we can delete the node holding the DWG Gray. Then, go to the Custom Curves and start drawing contrast down on the low end of the image. I often open things up a touch at the very bottom.

This is just my aesthetic preference. My way of doing things is not more or less correct than any other way. This method is just something I like to do by feel and by eye to give me an image that I like looking at.

Log and the human eye

Adjusting contrast is something I only ever want to do in log. If I tried to do the previous example in a linear space, my working range would be much smaller. The response of my curve tool would also feel weird, touchy, and difficult to work with. In general, I would have a tough time adjusting contrast in linear because things would not feel as perceptually aligned and intuitive as they do in a log space. 

Log space is much more aligned with the human eye than linear space. That’s because human eyes work on a log curve rather than a linear one. You can think of linear space as a modeling of physical light. It’s what happens out in the real world when things get brighter or darker. Log spaces, on the other hand, work more like our eyes do. For example, if I were to double the amount of light in the room I’m sitting in, I would not really perceive it as a doubling. I would perceive something less.

The meter on the left responds in a log fashion, and the meter on the right responds linearly.

A good way to think about it is that log spaces are good at modeling things that are perceptual and intuitive, such as preferential contrast adjustments. Linear responds in a more overall way, which we’ll look at later.

Look development in log

Look development is all about the preferential aesthetic manipulations we discussed in our previous example. None of these aesthetic choices are right or wrong, they’re just based on what we want to see, and they are going to feel more intuitive and easier to accomplish in log because log space is more aligned with how our eyes work.

Let’s move on to shot number two, which is another great example of how well look development works in log. 

I already have a look in place in this timeline, but I want to cool off the shadows. To do that, prepend a serial node at the timeline level and follow the steps we used in the previous example to get a DWG anchor point: Go to the Custom Curves, grab the DWG Gray patch, create a new serial node, and use the eyedropper to set an anchor point.

Next, go to the upstream node, un-gang the color channels, and select only the red channel. Grab the red curve and drop it down on the lower end of the Custom Curve. 

Doing this causes the image to lose some red. This cools off the shadows, effectively making them a little more cyan.

Now our shadows are a little cooler.
Now our shadows are a little cooler.

This is a preferential perceptual adjustment that makes sense and feels intuitive and easy in log. It would not feel nearly as good if we had done it in a linear space. Look development is all about aesthetics and preferential adjustments, so it works better in log. An adjustment like this example would be very difficult to achieve in a linear space.

Playing lift and gain against each other

Let’s move on to example number three. The method I’m about to show you is one of the first techniques that I learned when I began color grading. It’s all about playing lift and gain against each other.

Let’s start in our Primaries. First, move the gain northwest to warm up the image. Next, move the lift southeast  to get warmer shadows and cooler highlights. This is something I do based on what I like. I’m just feeling my way to a result that looks good to me on screen.

I wouldn’t be able to work this way in linear space because of the way the math works. That makes lift and gain another great example of something we would really only do in a log space. This method, along with look development and manipulating contrast, are all perceptual, intuitive, feel-based things that just work better in a log space.

When log doesn’t work

I love working in log spaces. There’s a reason that it’s the default domain for color grading. But there are some things that log isn’t good for. Let’s move on to the next shot and talk about linear space.

Here’s a shot that needs to be opened up. The exposure is kind of low, right? 

If we work in log, what are we going to use to open up this image? What is the closest analog to exposure available to us if we’re working in a log space? That’s right, it’s the offset. 

So let’s go in and open up the offset using the subject’s skin to find a level we’re happy with. Skin is generally what I look at when I’m making almost any kind of adjustment. First and foremost, I always want to make sure that I’m making my subject look great. But look at what happens when we’re grading in log. 

There’s a point where our shadows start to milk out, right? You can see it clearly in the histogram below. After the offset adjustment, there’s a big gap between zero and the bottom of the image. This is happening because the response of the Offset control in log space is not linear. It is not uniform up and down the tone scale.

The problem is that we are getting a faster response in the shadows than in the midtones or highlights. As a result, we would need to fix the image by going back in and dragging the lift back down, which might cause me to need to bring my offset back up even further.

Now we are in a dynamic that I try hard to avoid in color grading, which is making adjustments that demand compensations. Sometimes that situation is inevitable. But you’ll be a lot closer to the finish line if you can make an adjustment that creates no need for further compensation. 

Using the HDR Palette

Let’s use another method to compare the differences between log and linear. We’ll do this by using the Global color wheel in the HDR palette because that tool applies linear gain to the image. It’s actually doing the same thing as setting the Gamma on a serial node to Linear and using the Primaries to apply gain.

The only difference is in the units of measurement the Global wheel uses. The Global wheel uses photometric stops, which is a nice benefit of working in HDR zones. You can use either method, but for this article, I am going to use the Global wheel.

Working with exposure in linear space

Look how much less fogging there is in the linear space version of this image versus the log version. That happens because the response in linear space is more photometric. It gives a uniform response across shadows, midtones, and highlights. Exposure adjustments are a great hands-on example of where linear space shines because we get a cleaner, more uniform response right out of the box. And there’s going to be a lot less compensation when we use the linear Global wheel adjustment. 

Now, we may still want to make contrast adjustments, but they are not going to be made because something went wrong. They will just be the additional adjustments that we want to make because we’ve optimized our exposure. I’m not saying that you will never need to do anything else to your image just because you are working in linear space. But you don’t want to work in a space that is going to create more problems for you to solve. 

Color balance in linear space

So we’ve covered why exposure works nicely in linear. It’s more photometric, and it eliminates those compensation counter-moves that we would have to do right after making the initial move. That same idea is going to apply to color balance. Let’s take a look at another image.

This image needs to cool off. It’s coming in a little bit warm. Let’s start by using our Offset wheel in our Primaries Color Wheels to see a log adjustment. We want to find a much cooler rendering of this image. Again, I’m primarily using skin tone as a source of truth for how well I’m doing the job.

The skin tone is feeling good right there. Now, let’s compare our log color balance adjustment to one done in linear using the Global wheel in the HDR palette. 

This is a subtle example, but in log, the shadows are cooling faster than the midtones and highlights. It’s the same principle that made the shadows milky in our exposure adjustment example. When we make a balancing adjustment in log to cool the image off, the cooling happens quicker in the shadows.

The non-uniformity and non-linearity of the log adjustment create a need for counter adjustments. But that doesn’t happen in linear, where there’s a uniform response across shadows, midtones, and highlights. 

Wrapping up

This article just scratches the surface, but it covers the basics of log versus linear that I want every colorist to know when they’re going into a color grade. Log works well for perceptual field-based aesthetic adjustments, and linear works for corrective, absolute, physical things. In particular, I want you to watch out for log adjustments that are better done in linear because, in linear, there is less need to make compensating counter moves after the fact.

That’s really where linear shines. Linear gives us the right tool to make the right adjustment with no unwanted side effects. Then we can move on and look at other aspects of our grade once we’ve got the basics dialed in. Log helps us make aesthetic, feel-based look decisions because it works in the same logarithmic fashion as our eyes do.

Cullen Kelly

Cullen Kelly is a Los Angeles-based senior colorist with credits spanning film, television, and commercials, for clients and outlets including Netflix, HBO, Hulu, Microsoft, McDonald’s, and Sephora. With a background in image science as well as the arts, he’s passionate about the intersection of the creative and technical, and its role in great visual storytelling. In addition to his grading work, Cullen is an educator and proven thought leader, with platforms including his podcast The Color Code as well as his YouTube channel.

Insider Tips: Resolve’s Active Playhead

How a Pro Colorist Maximizes Color Separation

Should You Use Resolve Color Management or CSTs?