Original subtitles

When we looked at path tracing, we skipped over one major step in the creation of the

final render.

We covered how rays bounce around the scene, and how each ray will contribute some amount

of light to a certain pixel.

By the end of this process, we know how much light reached each pixel, but this isn't actually

enough to give us a usable image.

The issue is that the amount of light that reached each pixel can be anywhere between

zero, meaning no light at all, up to an unlimited amount of light.

Meanwhile, a screen can only emit light within a certain range.

So we need some way to map the unbounded amount of light for each pixel to a bounded image

color that can be displayed on a screen.

Perhaps the simplest way is to set a limit, and just clamp all the light above that limit.

This way, all the values within the range of the screen correspond exactly to an equivalent

light intensity.

If we plot the clipped values, it looks something like this.

As we'll see, this is a pretty terrible approach.

Let's first take a look at this black and white scene.

Here we can see that the sky reflected on the sphere is all clipped, and we can't see

any detail there.

We also see the reflection of the sun on the plane, which has a harsh edge where it starts

clipping.

This is not how our eyes perceive bright spots in a scene, there is no clearly defined boundary

where our eyes start clipping, in reality there is a much softer transition.

If we increase the brightness of the picture a bit, all the issues get even worse, now

we also have this harsh edge around the ambient occlusion under the sphere, where the plane

started clipping.

If we increase the brightness a bit more, the cone also gets a hard clipped edge.

And if we keep increasing the brightness, these clipped zones keep growing until most

of the picture is white.

Note that with each increase in brightness, the picture didn't really look like it was

getting brighter in a natural way, rather, it just looks like the clamped areas are growing

and flooding the image.

If we introduce color, the issues are even worse.

We still have the clamped sky and the sun reflection on the plane, but now there is

already a hard edge on the cone.

If we look at just the Red channel, we see that it's because it reflects so much red,

that the Red channel already clipped, meanwhile, the green and blue channels are still in range.

Here on the Blue channel it's also notable that the ground is already clipped, and the

cone doesn't reflect any blue light at all.

Now if we increase the brightness a bit, we see that the plane starts shifting to cyan,

because the Blue channel couldn't increase in brightness anymore as it was already clipped,

meanwhile, the Green channel kept getting brighter, so it changed the color ratios,

and shifted the hue.

The cone, and the leaves in the background also started shifting to yellow for the same

reason.

If we increase the brightness again, we see that the cone and the leaves clipped completely

to yellow.

At this point, the ground has clipped on all channels, but there is still this strange

harsh sliver of blue at the dark parts under the sphere.

And if we really bump up the brightness, almost everything goes to white, except the cone,

and the leaves in the back.

That's because they don't contain any blue, which means that with the red and green channels

clipped, we get this harsh yellow, as the Blue channel remains on zero, no matter how

much we increase the brightness.

The problem is that this is not at all how our eyes perceive brightness.

When we see something very bright, even if it's saturated at some specific hue, it starts

looking white.

Note that I'm showing these very extreme examples, just to illustrate the issue, but this problem

is still present even with much less bright images, and images can often also have a wide

dynamic range, with some parts being very bright and others very dark.

So it's important to properly convert the light intensity values from the render into

colors, in a way that's closer to how our eyes perceive the world.

To deal with the hard clipping issue, we can come up with a function where the output starts

off following roughly along the input values, but which slows the output down, as the inputs

get higher.

The idea is not necessarily to avoid clipping altogether, but rather to slowly ease into

it.

If we plot this function, we see that it has a softer high end.

But this function is only part of the solution, we'll also need to desaturate colors, as the

light intensity goes up, so that it clips to white, and not some harsh yellow or cyan.

If we run our brightness values through this function, and look at the black and white

example, the first thing we notice is how we have a wider usable dynamic range, and

we can even see details in the sky.

The reflection of the sun on the plane also has a much softer falloff.

Now if we bump up the brightness, it just looks like a brighter version of the image,

no hard edges, or noticeable flooding of clipped areas.

And we can keep increasing the brightness, and it still looks reasonable.

There can still be clipping, be we don't see the clipped boundaries, as we're softly rolling

into it.

Now looking at the color version, the cone doesn't have that hard red edge.

And if we increase the brightness, the plane starts getting desaturated, just like our

eyes would do, rather than shifting completely into cyan.

If we increase the brightness again, the cone starts losing saturation as well, and we still

don't see any hard clipped edges.

And if we really push the brightness, we see that we shift everything into white, even

the cone, which didn't contain any blue.

This looked much better, but there is a lot of maths and color science involved.

But luckily, we don't have to implement this stuff ourselves, Blender has this color transformation

that we looked at built-in.

It's applied by selecting Filmic in the color management options, but you don't even need

to do that, as it's already the default.

It's just important to be aware of this.

When you output a color, for instance, with an emission shader it'll look a bit different

in the render due to the color transformation.

When doing any sort of 3D rendering, you almost always want to have a color transformation

like Filmic being applied to your output.

The only case I can think of where you wouldn't want such a transformation, is when doing

flat rendering for something like motion graphics, where all the shaders in the scene are flat

emission shaders, and you want to control the output colors directly, in which case

you'd select the Standard output transform, which just clamps the values.

But as soon as you are applying any sort of lighting to your scene, you'll want to have

a proper color transformation applied.

This was an overview of color transformations, and how we get a proper color from a light

intensity value.

But we are barely scratching the surface of the color science required to actually make

an image.

If you want to have more understanding of the whole color pipeline, and how colors get

from a source all the way to the displayed output, I can highly recommend The Hitchhiker's

Guide to Digital Colour, which incidentally is written by Troy Sobotka, who is also the

author of the Filmic color transformation.