All language subtitles for 02. Discovering deep compositing basics

hi in our first lesson

I will be reviewing the basics of deep compositing

for viewers with no previous deep experience

here I have a simple deep image

and this image is composed of two different parts

it's composed of

the motion blurred red sphere element

as well as the blue cylinders element

now you might be asking

what is deep compositing and how does it work

so in a normal CG render

every pixel in the image gets one sample

one sample of color

one sample of depth

one sample of every other AOV

recorded in the output file for that pixel

the difference is with deep

you get multiple samples per pixel

so what happens is

the CG scene is divided up into multiple depth slices

and each slice gets its own sample recorded

and I can show you this

using the deep sample note

as I search around

you can see that

every pixel has a different set of samples

now it's important to understand that

deep sampling is adaptive

it's not constant

so in situations where

for instance

we have one sample per pixel

that's because there is nothing behind this pixel

there is no other cylinder behind this

so only one sample is required

to accurately describe the depth of this pixel

when we get into

layering where there's multiple cylinders

then we get more samples

here at this edge between

the foreground cylinder and the midground cylinder

we see there's two samples

we get into things like the sphere

we get many more samples because it's motion blurt

has a lot of transparency or opacity

and it requires a lot more samples

to accurately describe its depth

here we get 106 samples at the single pixel

so I can visualize this for you a different way

I can visualize it in Nuke's 3D space

using the deep 2 points node

I can generate a point cloud

so here we see

a 3D point cloud

reconstruction of the 3D scene from Maya

and we can see that the cylinders behind this sphere

this is the camera over here

so from the camera's perspective

which is about right here

there is still data for these cylinders

now when we compare that to

the kind of data that gets recorded in something like

a depth pass

we can see where there's only one sample per pixel

where the sphere excludes the cylinder

there's no more information it's gone

so what does this allow us to do

but we can do really interesting things

what we can do

for example

is we can actually move the objects around

normally this isn't possible

and it's not possible because

in a normal workflow

with one sample per pixel

using a beauty Pass and an alpha channel

and a depth pass and all the other traditional AOVs

you need to render holdouts

so if this fear is animated

it would go through multiple iterations

you'd have to re render

and every time you re render with a new animation

you would have to render holdouts because

if all you have for compositing is an alpha channel

you need to have alpha channels that don't overlap

so it means that

for these pixels

the alpha channel of the cylinders

would need to become progressively

more and more transparent

and then eventually disappear completely

and the inverse would happen to correspond

for transparency in the sphere element

so you would have holdouts for the animation

baked into the CG renders with deep

the holdouts

are generated in compositing

that holdout process is deferred to compositing

so the CG render is rendered as a whole

see the cylinders are whole there's no holdouts

the sphere no holdouts

but when I combine them together

they're accurately merged

that's because of the deep data

the deep data is describing how they should intersect

now something interesting about this is that

because we have a complete

volumetric sample of these two different elements

I can actually change their relationship

and accurately

reconstruct what the elements would look like

in a different relationship

so for example

I can put this behind all the cylinders

I can push it back

and you can see as I as it moves through through space

it's as if I'm natively rendering this

it looks like it would if it was just rendered this way

zoom in a little bit

so we have a better view of this as we go through

this normally isn't possible

but we can do these kinds of things in compositing

because of the deep samples

now at this point you might be thinking

I know what a depth pass looks like

I know what a position pass looks like

but what is a deep pass look like

so if we go to the viewer

who go to the deep layer

can see here

this is deep

so why is it yellow

it's yellow because it only has two channels

red and green

and the values in this particular case

are all super white

they're all above 1

so what do we see

we see a clipped red and green channel

which gives us yellow

solid yellow

now

as I move my cursor around

I can see the values are changing

and of course we know the values change in deep

because we use the deep sample

can see here

move around

we see the values

over here the values change on every pixel

so

if I expose down

in this case to F2 56

we see something that looks more familiar

we see the cylinders

we see the sphere

and this is just raw deep data

and there's no other way to really view it

in the viewer as a 2D thing it's just

it's just what it is and it's a

it's a technical pass

it's not really meant to be looked at

and at no point do you really ever need to look at it

there's nothing you really learn from looking at it

I just want to show it to you

so you kind of understand how nuke interprets it

but deep itself as a

as a visual thing is

is sort of abstract

now two channels

I mentioned that there was multiple samples per pixel

but why are there two channels

the reason is because every sample

per slice actually has a front and a back

so as we go through here we see at this Pixel

there are front and back

there are front and back slices so

can see all 32

and if we look at the difference here

we look at the red channel this is the front

look at the green channel this is the back

so that's what deep looks like

that's why it's got two channels

the last and one of the most important things to cover

is deep nodes

now deep data is

a kind of data that's only compatible with deep nodes

you can't use regular nuke nodes with deep data

you can think about deep as something like

the way you would use Nuke's 3D nodes

or Nuke's particle nodes

they're not

you can't just plug in a regular 2D node at any place

anytime and expect it to work

it's because

the kind of data that's flowing through the pipe

is not the kind of data that the other nodes

can process

so for instance

let's say I wanted to take this image

and blur it

create a blur node

try to plug it in

it won't connect

there are two nodes that

you would use on a regular basis that will connect

and they are the dot

and the no op

the reason these two nodes work with deep

is because they're just pass throughs

they don't perform any processing or functions at all

they're really just visual aids on our node graph

so I'll show you in later courses

how to deal with deep nodes

how to deal with this restriction

and it's a pretty severe restriction

is really a very limited number of deep nodes

and I'll show you when

it's advantageous to use deep nodes

and when it's more advantageous to convert to

a regular 2D image and work with your comp that way

so I hope you stick with me

and I'll see you in the next course