Now, previously we saw Gamma workflow and how it can produce natural looking results. So then is there a problem with Gamma workflow? To put it simple while Gamma gives visually pleasing results, it is not Mathematically accurate. We mentioned before Predictable vs Accurate. **Please note that every software and hardware works linearly under the hood.**

8-bit. Two spots. Halved values. Total is 255 or 1.0 or %100. Now to the point where the problem arises.

Common simple Mathematical operation 2+2 **≠** 10. What does it have to do with our topic? Let's see Color Space graph:

Yellow curve - sRGB-Gamma 2.2 transformation.

White diagonal line = Linear - Gamma 1.0

Here we see how the math breaks down:

-sRGB midtones, 0.5, %50 equals to 0.2 in Linear gamma.

-In normal math sRGB highlights can not even match to Linear midtones! (0.2+0.2=0.4)

-Due to curve exponentiation of sRGB, both highlights equalized.

Back to the spots. We saw that two halved sRGB combined in Linear is equal to 0.4 or %40. Despite that we see %100 as a result. How sRGB 1 becomes Linear 1?

As we mentioned in our previous article, Gamma steals, trims bright bits and transfers some of them to the darker bits. Back to the graph:

-In normal math, sRGB highlights should be equal to 0.4 or %40 on Linear.

- We have a deficit of %60 on Linear side.

-sRGB curve stretches from %40 to %100 on Linear in order to equalize shade steps.

-Equal range means less and less brightness.

-Resulting predictable but inaccurate image.

As we mentioned in Color Depth, Luminance is everything. More steps and bits in bright area means more room for adjusting image without washing it out and more importantly we will have Mathematically accurate workflow. When we switch to the Linear, we will have a lot more bits in range, without trimming bright or dark.

Let's see sRGB problem and Linear solution in digital production:

An sRGB ramp. Left hand %100 Red. Right hand %100 Green. Middle? Forget about %200, forget about %100 which is max output of our screen. Middle brightness is reduced all the way down to the around %30 percent. Rings any bells? Back to the Color Space graph to explain further:

sRGB %50 equals to Linear %20. 20+20=%40. You can not curve Math!

If we switch to LWF we can fix it for good:

If we make the ramp radial, the problem and the solution become even more obvious;

This solution will return to you as both Math accurate and visual please in every software, FX you use.

Let's see a similar common problem, the root cause. Non proper working Add blend mode.

So as per Math, two spots of 0.7 will result to 1.4.

If we don't linearize sRGB, it will not go beyond 1 or %100.

If sRGB can not go beyond, beyond will be squeezed into sRGB (that 0.4 above 1)

This will result overexposed highlights in blend modes.

We blended fire with Add mode. Since we don't see original fire here it'is impossible to see difference now but notice how fire blended brightly. When we switch to the Linear we'll see the truth. So why fire washed away the image?

Again, Color Space:

Inspect graph over two points:

sRGB Highlight - Linear Highlight = 1

sRGB Midtones = 0.5 = Linear Mid 0.2.

So while Highlight should be 2x Midtone, it is 5x in linear. Put it opposite way, sRGB curve should be extended around value of 1.5!

Since it is not possible to extend, that excess 0.5 will be squeezed into the 0-1 range.

Which will result overexposing.

If we linearize it we can clearly see the difference:

Linear Additive

non-Linear Additive

Since LWF is the core of softwares, every FX will result way more deep and quality image:

Altered Carbon :

Non-Linear Glow effect. Notice how bright parts washed out:

Linear Workspace. No washing out. More natural Glow effect:

Notice how glow extends further without washing image. Like atmospheric haze:

Another example, commonly used Post FX, DOF. See it with FL Out of Focus:

Not looking bad but bokeh should be more vivid and sharp.

Linear Workspace:

In next article we will see LWF based on softwares and how it works depending on them.

Stay safe.

## Comments