All language subtitles for nuk231-class09

hello and welcome back to nuke 2 3 1

nuke composity best practices

this is Victor Perez

and in this class

which is class No. 9

we are going to start

talking about optimization of the 3D renders

so we're going to talk about this can line render note

okay so let's start by one of the most obvious things

which is the format of the skyline render

okay you know that the skyline render has three inputs

one for the object or the scene

the camera and then there is the mysterious input

of the background well

make no mistake here

because the background is not defined

to put the actual background of the composition

even if you can

then if you do that

you cannot separate

the object you are rendering from the background

so we used to put

ah reformat

so the only information this background input

is going to grab from this reformat node

is the actual format

just to output my image in this format

what if you use another note well

there is many people that

instead of using the reformat

they prefer to use the constant

but this is wrong why

look at the boundary box

in theory this reformat

and this constant have the same information

but it's not true

because in the reformat

you get the format of the image with nothing inside

and for nothing

I mean there is no information

as you can see here

I'm just grabbing the information with the color picker

as you can see

but here there is nothing

while in the constant

when you pick

in here you have information

but that information is zero

but it's not the same

have no information

to have information

equals to 0 in every pixel

because in here

your boundary box

is going to process every single pixel

and to be honest with you

nuke doesn't care

if that information is equals to 0

or equals to 2000

is the same

that this information is going to be calculated

so when you are applying the format

to the scanline render

never use a constant

use a reformat instead

okay so that is going to limit the process

of your scan line render

which is an old

that sadly is quite slow to render

and is going to limit that

to this boundary box

which is defining the geometry

okay so remember that the boundary box

in a projection

for instance

is not matching the alpha

or the boundary box

of the image you are projecting

the boundary marks from the scanline render

is always going to match

the geometry

that is going to be rendered out

okay so if you see here

the same image

has no boundary box

is all to the format

okay so another thing is

why not to put anything in there

so leave the input open

well if you do that

and then by any chance

you copy and paste this note

to another project

that is using another project

settings for the full format size

then your skyline render is going to change

and everything else is gonna change

so it's not a good idea

to leave the background input open

you need to put something

and when you put the reformat like here

don't leave the reformat

at the root format

instead use a format in here

so by default

the root format

is going to take

the format of your Priyat

so if your Priyat is set

like you here

at 2K Super 35

for aperture yes

put it in here

don't leave it

to the root

because if you put a reformat

and then you leave

to the root

is exactly the same as putting

nothing okay

so you have to specify

the format in here okay

so that is going to create

a right format

that is going to

be perfectly fit

with the background plate

and then you just need to put a merge

and put this over

or any other operation of your choice okay

now another thing that is very interesting

when we are working with the skyline render

is sometimes we have an image like this okay um

maybe you want to apply a certain process to this image

okay that process can be

for instance

a color correction

or a convolution

for creating

for instance

a glow or any other process you want to apply

so how to do that

well depending on the operation

you want to use one method or another

okay so imagine that I want to create

some kind of color correction

okay so if I have this

and this is my alpha

there is no way you can separate that

but remember

the scanning render is going to render every channel

you are putting inside the UVs of your geometry

saying with the predictions okay

so if you having here a prediction of gas r g b a

the output result is going to be RGBA

but if I want to isolate

for instance

this first sphere okay

that is corresponding to this okay

what I want is just to create kind of ID pass remember

the ID pass is the pass that is going to give me a mat

for a certain element or geometry

okay so how to create that

well it's going to be really

really easy

I'm going to call a shuffle note

and then I'm going to place that shuffle note in here

and in this case in

instead of taking any pass from my original image

and then placing into the RGB

what I'm going to do is

I'm going to take one output layer of my choice

that can be

for instance

the mask or

if you prefer

you can create it yourself

so in this case

I'm going to put the mask

because I just want one single channel

and now I'm going to feel that mask with full 1 value

so if you look at this

this is RGB

now this is the a

okay and now

if I go in here

the mask here you are

this is the information of my mask

that is display

by the way in red

just because it's the only channel available

but it's not that

my mask is red

is I'm just caring about one single channel

and the first channel available in nuke

he's the red one

so that's why the aparency of my mask is red

so let's go back to the RGBA set

okay and now

if I see my output

you will see our GBA

okay but attention

what is happening in here

now I have an extra channel

okay that extra channel is the mask

but he's being applied only to this geometry

so if I go here

to the mask

here you are

so now that mask is isolated

so if I want to apply a certain effect

I just need to go here

great now I'm going to use the mask

okay so now I'm going to create that certain effect

for that piece of geometry okay

but that is actually not perfectly accurate

when I want to create a color correction

can you see what is happening in here

the edge is creating that glowy effect

and that's because I'm using a mask

that is matching the alpha

so it's exactly the same effect of applying

instead of a mask for the ID

I'm going to apply the mask for the alpha

so now the same effect is happening for everything

and as you know

you never use the alpha as a mask

the alpha must be used as a pre multiplication process

okay so you unpro multiply

you then apply the process

and then you pre multiply

this is something that you cannot accomplish

with this process

because if you use the mask for um

primitiplication

something like here the mask

then everything is going to disappear

because he's applying a non primal duplication

and then the primal duplication by this shape

which is just creating that effect

and even if you force that

and you for instance

instead of doing that automatically inside the node

you prefer to do it manually with nodes

so I'm going to call it here

the amprimoult

so I'm going to amprimoult by the mask

and then to premoult after by the

alpha okay so in here

you get that

and in here

everything looks fine

except those edges

and those edges

so there is no way you can change the color

with that extreme color correction

by using a mask

that mask can be used very nicely for a glow

or for delimitating

any effect of convolutions that it is okay

or to mask other

effects that you can apply to the other spheres

like set blur or something like that

actually this method of isolation

is not great for color correction

okay so how do you do that

in case you want to isolate an element well

that is nothing better than getting your sphere

separate in a different render

but in this case

is going to be really easy

if you apply this fear to be the first fear

so everything is going to be under this fear

in a layer combination okay

but what if this fear I want to change is the green one

well things can get complicated

okay so I'm going to show you how to integrate

deep image compositing to apply that certain effect

so in here I have

the render of my red and the blue spheres

and here I have the render for the green sphere okay

you know that automatically

even if you don't want to

this calendar is releasing the deep information okay

so if you use something as simple as the deep merch

here you are

you are going to get this merch

and you are going to mix this and this

and together

here you are

okay so he's going to respect what is in front

of my green sphere and what is beyond my green sphere

that's very nice

the problem is

as soon as you put a node after this can line render

the information

of the deep image compositing is going to get lost

so how can you manipulate maybe your green sphere

and still maintain this deep image compositing setup

well for that

we need to get a little work around

using another note of deep image compositing

which is in this case

the deep holdout

so the deep holdout is going to create the hole in here

just to place this green sphere

using an operation that is not the over

okay every time we are using a holdout

we are going to use the

this joint over

I'm going to show you now

but first let's apply the deep holdout

so in here you are going to apply the image

that you want to crop from this image

okay so that is the holdout input

and then because this is my main element

that is going to be feed into the main input

remember that the holdout is a note

that is a processing deep image

but the result

the output is going to be a regular to the image okay

that's why you are getting in this note

the rectangles of the color

so every note that has the rectangles of the colors

corresponding the channels

that is a to the process

okay so in here

if I apply this to this

the result is going to be that

so is the portion of the sphere

that is reveal by the rest of the scene

so the red sphere in front of the green

is now cropping those pixels okay

um before to apply a glow

because this is very simple

you apply a glow

and then you put in everything together

I want you to think about something

remember that we are applying

a mask to crop those pixels

okay so if we put anover

the effect that we are using is we have that

and this

of course in this schematic drawing is very clear

in reality this is happening exactly the same

which is this area is not touching

well it's actually very very

very similar in the value

so if you use an over

probably you're not going to notice much

but as soon as you are getting motion blur

or soft edges

this portion of the edge of the alpha

not being perfectly match

with the image above and below

is going to create a dark edge

okay so if you don't want to have that dark edge

is better that you don't use and over

because the over is designed to put two images

that are just a main image with transparency

over a full cover image

but in here

we have already my image pre cut

so if I put this and this

to deactivate the glow

and to show you

have a look at those pixels

this is kind of pinky

and this is kind of green

so in here we have this tiny

tiny variation in the value that is getting dark

which is because in there you have actually dark values

and in here

you have the same dark value

so when you put those pixels together

you don't get the full amount of value

just to represent the perfect blend

from the pink to the green

so when you have this effect

you have to use the this join over

okay the this join over represents

when the alpha is not perfectly feeling all the gaps

so with the this join over

you're going to get this perfect sense of blending

of an image with the holdout

okay so with this drainover

everything is looking great

so I'm going to check if this and this

which is the deep image

with the merch

and these with the this join

are accurate

okay so here you are

perfectly accurate

from one to the other

okay notice that in here

if I change

and I compare the over and the deep match see

this is the edge I was talking about okay

now that we have that merge perfectly mixing my images

now I can apply the process

and if you want to apply the process

you can apply the glow

and everything is going to work fine

even the pixels that are getting greenish

on the pinky area

but not only in here

I can also apply a great

and in this case

because this image is just a pre multiply image

I can apply in here

yes I'm promote remote from the note

and then I can change anything

and is going to look perfect in every single pixel

okay so this way

using the deep image compositing

is going to speed up your overflow

and is going to avoid you

to create three layers of the skyline render

one from the red sphere

one for the green sphere

and one for the blue sphere

and in this case

is really easy

because they are not touching from one to the other

but when you have things that are going

one in front of the other

and then back behind

then everything is going to be a bit messy

so using deep image compositing

for this kind of holding out

is going to help a lot okay

so that is for layering

using advanced techniques of this calendar

another thing

that we have to keep in consideration is that

when you are rendering an image with motion blur

everybody knows that the motion blur

is actually being created by samples

so I'm going to show you

in the skyline render how the samples work

so here we have the multi sample

so the amount of motion blur

and the amount

I mean how long is the tail of your motion blur

is going to depend in terms of quality

of the amount of samples you are going to apply in here

because the samples are just a render of the same image

in different position of times

in between the frame

so is rendering from frame 1 to frame 2

several different positions in time

so let's say

render 1.1 frame 1.2

frames 1.3 and so on

it's something that is going to

blend in one single frame

so you are going to have

the feeling that that image is actually moving

but is related with the amount of samples

and the amount of samples is as plain as

you are rendering this image 100 times

in this case

if you render twice

that is going to take twice the time for rendering

if you put 5

is going to take five times

as you can see

it's really clear that if you put 3

you can see the shape of my image three times okay

and he's blending everything together

but to get a good effect

depending on the amount of mushroom blood

you are going to need several samples

so in here I decide to put 100

some of you can think now okay

if I render something complex 100 times

that is going to take a lot of time for calculation

and that's true

so then in here

if you don't want to calculate the samples

you can calculate what is called

a motion vector pass

and in nuke

you have different kind of motion vectors

depending on the information you want to represent

okay by default

you are going to get distance okay

which is the distant

that every pixel is representing movement

in this case

forward okay

from one pixel to another

so then once you have that pass

which is looking something like this okay

is representing in u v

which is the relative coordinates x y

he's representing

the distance from this pixel to the next one

intense of movement

so using that information

which is baked in the render

pixel by pixel

I can use what is called a vector blur

the vector blur is in here

and it's going to apply that

a movement of every pixel

leaving the trail

from the original position

to the position of destination

in both X and Y

but keep this in mind

is just X and y

there is noted in here

so you cannot create 3D motion blur in here even

because what you are doing here

is just a compolution of pixels

based on a matrics of movement

but is just

that is very different from getting the image render

in several different positions in time

between one frame to another

in here is just plain movement into the

so when you apply those motion vectors to the image

you get something like that

which is more or less similar to the final image okay

as you can see

there are artifacts

and that's normal okay

with the vector blur

I apply a certain channel

then I use my alpha to define

what part of my image needs to be moved okay

and then I was applying the amplitude of my movement

so I don't care about the setup of this vector blur

what I care is about the process itself

and the process

let me tell you that

calculating the motion blur

using the vector blur is not a great

sometimes you can be lucky and don't get

those artifacts

but most of the time is kind of dangerous to apply that

and even if the result is quite similar

the quality of the motion blur from the 3D

and the vector blur is so different

there is no way you can even compare the result in here

is is just a process of today

but your brain is going to be very smart

and is going to understand

when that motion blur is real or not

even if in the skyline render

the motion blur that we are calculating

is just a linear fashion motion blur

so it's a mathematical way of representing

the soup information from the time

while in here

you have just

in this case

24 segments of time

but we are inventing

the information that is in between those frames

the vector blur is not a good idea

just try it

but just a tiny amount

because the vector blur

he's not going to be used

in general I like to use the vector blur

and to apply the motion blur

when I'm applying a patch over a CG

and then I need to

create the movement of that patch in the CG

with the same

more or less

fashion of the movement okay

but by default

I would never recommend you to use

the vector blur for general motion blur okay

still if you want to get better

results with your motion blur

I'm going to give you an advice

which is in here

you have this randomized time

I'm going to reduce

my samples by 20

so you are going to see clearly

the pattern

of the linear representation

I mean linear

because the distance between all samples

is arapematically the same

okay so in here

everything is looking like

mathematically perfect

but you can apply

what is called a randomization of the time

so is going to take

different distance

for every sample

so that is going to hide

the effect of Muay

for instance

which is those lines

doing weird things

when they duplicate

like in here

and that is going to hide many problems

of course you are not going to resolve

hold the problems of the samples

but my aiming here

is how to provide you the information

to render your image

to look fine

without having a crazy amount of samples

so if we randomize the time

we can create that irregularity

okay so in here

you can see very clear in those lines

that the space between the lines

is not the same as when you were in here

see that everything

all the distances are quite the same

okay so if I increasing here

and I put one that is going to be randomized

and that this is starting to help

of course he's not perfect

but he's starting to help

then another thing that can help

is the use of the stochastic samples

this torcastic sample

is going to render more information

where you can see it

and less information in the areas that are not

so important

and we are talking in here about perception

so nuke is deciding where to use more information

where the information is more dense

and then applying less information

that is going to create an effect of noise

because it's going to create clusters of information

so we are going to increase by zero

you are getting just nothing

but as soon as you put one

you are going to start getting information in there

one is very

very very low

so in this case

as you can see here

I have a good density of points

and then in here

that's not much

see that there is nothing in there

it's just valued zero

okay of course

if you are at this distance

probably you are going to be more or less fine

but it's not a good idea to use just one

so we are going to increase

up to the point that we can just get things to work

good for us

don't get confused

this is not grain or noise

this is that the result of getting the information

group in clusters of a perception okay

so with this difference is

is going to take way less

of course is related with the definition of my image

if you have a very

very noisy image

that is going to create this kind of noisy pattern

okay but in terms of the quality

you need to get the final result of your image

and then balance the stochastic samples as well

okay so pray

you get a good balance in there

okay then once you have your good balance in there

just try to apply the sample diameters

the sample diameter is related with the antialias

so is how big is the circle of pixels

just to create the amount of antialia

so the higher this is going to be

the more accurate antialia as you are going to get

that is for very precise calculation

in this case

because the motion blur is so high

then you are not going to get

that increase in the quality okay

the focus diameter

of course is related with the death of feel

of the camera that you set up

so I'm not going to talk about this

but this is going to be the effect of the set blur

done in camera

so as you can imagine

that said blur is just a combination

as the vector blur

but if you create the difocus

directly from the calendar

is going to work way better

okay so the focus diameter is just going to define

the amount of distribution of your samples okay

good so now

once I have my setup

I can increase

and see now

by using 50 samples

probably I'm going to get a quite decent result

and by combining the stochastic samples

and the amount of samples

and reducing the time of the render by a huge fraction

even if I was using the hundred samples

that I was using initially

the stocastic samples

are going to reduce the time of the render

because he's not using the full precision

of every single pixel for blending the samples

okay so the stocastic sample can be a very good asset

to reduce the render time okay

another thing that is interesting

when we are working with the scanline render

is to use your channels wisely

so in this case

for instance

I'm creating my motion blur

directly from my skyline render

which means that probably

I don't need my UV representing the motion vectors okay

but in here

by the fault

this calendar is providing me with the motion vectors

so what you want

is to render just the information you need

so here you go to this shader

and then motion vectors off

as you can see

here you are rendering to channel less

okay that's good

and then you can say okay

and what if I don't want my channel

because in here

I'm rendering also the death information well

there is no way you can remove that channel

even if you remove the set buffer or the transparency

that's not related

okay so if you want to get rid of that channel

you just go in here

remove and then you can either remove the deaf

okay so is bypassing the other channels that's away

or if you prefer

you keep just the RGBA

okay so he's going to discard

any information that you don't need in there

so that is going to create a very clean stream

lining here with just the channels you really want

okay and now we are going to discuss the boundary box

in relation with the lens distortion

so I have this image

with the camera that I provided you

and then in here

I have the render of my cones okay

so those cones are representing

certain areas of the 3D space okay

but in here

in order to match my virtual image

with the original plate

I need to get my lenses torsion align

and let me underline this point

you never touch the original plate

by using the and distort for compositing reasons

you use the and distortion

just for calculating your camera

when you are applying the much more process

but never never use this

which is from the calendar

then you apply the and distortion to the image

and then put both together because

yeah it's true

they are both align in terms of lenses torsion

but you are breaking in here

one of the principles of the capacity

which is you are touching

something that is not mean to be touched

which is the background plate

if you don't have a particular task

that requires that be touched

so you are losing a huge amount of quality

with the lens distortion

which is applying a filter

okay so never do that

what you want is just to apply

the lens distortion in here

just to read

distort your CG

in order to match the original plate

with the distortion

um this is happening always

there is no exception

all cameras in the world

have a certain amount of lens distortion

some of them

they have more

some of them have less

but there is always some

okay so if I apply this

to the process I did before with the stochastic samples

okay you will see that now

because of the filter I'm using here see that

you are getting a bit of the reduction of that crane

so you can play back with the stocastic samples

and the filtering here

so if I use something like the notch

see that now is reducing

because he's blending those samples okay

so this is something that you can play with

in case you need it

so let's get back in here

so I apply my lens distortion to this

and what is this

well this is totally normal

is because when you are applying

the lens distortion to your image

you are doing this

you are and distorting your image

by taking the corners and stretching them out

okay so the image seems to be enlarging

I'm going to show you in here by using a black outside

okay and then in here a note that I have in my toolbox

which is the box to format

so can you see what's happening in the edges

the image is getting a stretch in the corners

so if I change from there

see that the image is actually looking like

the center is getting back

but it's actually the other way around is the corners

so this is a non linear transformation of the image

okay but what is actually happening is that this image

that is getting outside

because we are stretching the corners

then when you apply redistort

those areas that are outside are getting back inside

okay but what if I apply this

and then I crop my edges so

I'm going to

cropping here okay

see what's going on

is because this crop is erasing all the information

that was in those corners

outside the image

so the first thing we cannot serve is

the boundary box is expanding okay

so what is happening to the scanline render

is exactly that

your render is just up to the format

so when you apply the redistort

those pixels that are supposed to be outside

with the boundary box are not there anymore

so this calendar is just rendering the format

and then in here

the format is getting inside

so you are getting the regular issue of the expansion

of the last pixel touching the boundary box

okay so how can we fix that

well there is a feature in the scan line render

that is called the overscan

so the overscan is just mean to expand

whatever is not seen by the camera outside the format

but is information that in a regular 3D virtual world

we can apply

so we can apply that

and you can feel how the boundary box is expanding

okay so when I move

for instance in here

I'm going to apply a transform

just to move

we are going to reveal a piece of the corn

that is not seen

as we did in the class talking about

the boundary box

see the cone is in there

so if I restored that

the information outside is going to be now insight

due to the calculations of the lens distortion

so now when I apply that to the image

that is going to appear correctly

the point is

how they overcome

should be well as big enough to cover that

so it's something that you are going to operate by eye

my advice is

when you are going to calculate the oberscan

go to the corners of the image

so in case your cones

are getting to the corner of the image

that is the most critical piece of the image

because remember that the lenses torsion

is actually stretching from the corners of the image

okay so there is going to be easier

and then add a few pixels more in the overscan okay

just to be sure you are not cropping

but be careful

because the higher is the overscan

the more pixels you are going to render

from your scan line render

so you just want to keep the scan line render

with less pixels to be render as possible

without sacrifice the information you need

then another thing that I used to do is I always

when I use the merch

put the information coming from the virtual world

with the original plate

usually I don't want to change

that information in terms of transformation

so what I do is just to simply crop my boundary box

and limit that to the B input

which is the original plate

so in here B

and now the boundary box

is perfectly fit with all the information inside

so I hope you enjoyed this

set of tricks and methods for working with the 3D world

okay so that's all for this week

if you have any comments

questions or suggestions

you are very welcome to post in the forums

until next time

this has been Victor Perez for effects PhD

thanks for watching