Would you like to inspect the original subtitles? These are the user uploaded subtitles that are being translated:
1
00:00:02,147 --> 00:00:05,147
In this clip we're going to take a look at luma keying in depth.
2
00:00:05,147 --> 00:00:09,320
Now in our previous clip we did take a look at the keyer and that included
3
00:00:09,320 --> 00:00:12,601
luma keying and a couple of other different types of keys.
4
00:00:12,601 --> 00:00:16,237
We looked at all the different parameters and how to
5
00:00:16,237 --> 00:00:19,146
modify them to really get the best key.
6
00:00:19,147 --> 00:00:22,447
Now in this clip we're going to look at practical examples.
7
00:00:22,447 --> 00:00:25,147
So let's take a look at what we have.
8
00:00:25,147 --> 00:00:28,328
So the first one is this image here which is a lighter image on
9
00:00:28,328 --> 00:00:31,813
a darker background and the second one is going to be a darker
10
00:00:31,813 --> 00:00:33,480
image on a lighter background.
11
00:00:33,480 --> 00:00:37,531
Now the workflows for both of these are included in the script,
12
00:00:37,531 --> 00:00:42,147
but we are going to be primarily focusing on our first image.
13
00:00:42,147 --> 00:00:43,973
So let's take a look at it.
14
00:00:43,973 --> 00:00:47,364
So here's our first image and we have denoised it just to get
15
00:00:47,364 --> 00:00:51,323
better edges so that that noise isn't involved and I pulled a
16
00:00:51,323 --> 00:00:54,146
really quick keyer so this is a keyer.
17
00:00:54,147 --> 00:00:59,385
I have modified the A, B, C, and D sliders.
18
00:00:59,385 --> 00:01:04,623
We can see that the backing region is uneven so
19
00:01:04,623 --> 00:01:07,546
unfortunately we can't do a whole lot with this.
20
00:01:07,546 --> 00:01:11,146
It's not going to allow us to push that key any further.
21
00:01:11,147 --> 00:01:16,099
The backing region here is in around that magical 5% so we have
22
00:01:16,099 --> 00:01:19,146
to modify our settings a little bit more.
23
00:01:19,147 --> 00:01:21,480
So what do we need to do?
24
00:01:21,480 --> 00:01:21,813
Well,
25
00:01:21,813 --> 00:01:26,146
we need to look at possibly evening out that backing region and also
26
00:01:26,147 --> 00:01:29,702
combining multiple different keys into a final matte.
27
00:01:29,702 --> 00:01:34,146
So let's take a look at some of those workflows.
28
00:01:34,147 --> 00:01:37,146
So back into our script and let's take a look at this section here.
29
00:01:37,147 --> 00:01:41,147
So this section here that is in the teal color,
30
00:01:41,147 --> 00:01:45,147
this is the part of the preprocessing that is slightly
31
00:01:45,147 --> 00:01:48,094
different than what I have previously shown.
32
00:01:48,094 --> 00:01:51,883
So if you are new to the preprocessing workflow,
33
00:01:51,883 --> 00:01:55,646
I do highly suggest that you check out the NUKE Green Screen
34
00:01:55,646 --> 00:01:58,647
Keying Fundamentals Course on Pluralsight and check out the module
35
00:01:58,647 --> 00:02:02,147
and clips that go through this workflow.
36
00:02:02,147 --> 00:02:07,147
So let's take a look at how I modified this for this particular example.
37
00:02:07,147 --> 00:02:11,283
So the first thing that you will find that is different is that
38
00:02:11,283 --> 00:02:14,146
this is a non-green screen or blue screen image.
39
00:02:14,147 --> 00:02:17,897
So if I take a look at this we're going to have to find another
40
00:02:17,897 --> 00:02:20,147
workflow to create a clean plate and a key.
41
00:02:20,147 --> 00:02:25,046
So I have taken a luma keyer and I have used a gray tool
42
00:02:25,046 --> 00:02:27,146
just to really optimize the alpha.
43
00:02:27,147 --> 00:02:29,432
So I have changed the channel to alpha,
44
00:02:29,432 --> 00:02:33,146
I have clamped the white value so that none of those alpha values
45
00:02:33,147 --> 00:02:38,347
will go above 1 and I've aggressively increased the gain so that I
46
00:02:38,347 --> 00:02:41,147
can really pump up that alpha channel.
47
00:02:41,147 --> 00:02:45,424
Now from there I did a slight dilate out with the
48
00:02:45,424 --> 00:02:48,146
filter erode and I inverted the matte.
49
00:02:48,147 --> 00:02:52,147
I then took the regular RGB image and premultiplied it.
50
00:02:52,147 --> 00:02:55,747
So again this is kind of an inverse premultiply
51
00:02:55,747 --> 00:02:57,813
from what you're probably used to.
52
00:02:57,813 --> 00:03:00,147
Now this is the little trick here.
53
00:03:00,147 --> 00:03:03,407
So if you take the edge blur and you blur it ever so slightly so
54
00:03:03,407 --> 00:03:05,499
maybe a little bit more than ever so slightly,
55
00:03:05,499 --> 00:03:11,146
but you're looking to take that blur all the way into the center of your image.
56
00:03:11,147 --> 00:03:16,304
When you un-premultiply that, what will happen is it will fill in the gaps,
57
00:03:16,304 --> 00:03:18,147
which creates a clean plate.
58
00:03:18,147 --> 00:03:21,897
So the idea being is that we don't change the
59
00:03:21,897 --> 00:03:27,147
original image as much as possible, but that we just fill in those gaps.
60
00:03:27,147 --> 00:03:30,747
So this is equivalent to the IBK color node that does a
61
00:03:30,747 --> 00:03:33,147
better job for green screen and blue screens,
62
00:03:33,147 --> 00:03:37,347
but this workflow serves as a really great alternative
63
00:03:37,347 --> 00:03:40,147
for non-green screens or blue screens.
64
00:03:40,147 --> 00:03:43,447
So the next one is pulling that difference keyer so that's
65
00:03:43,447 --> 00:03:47,459
equivalent to the IBKGizmo and we hook up the difference tool up
66
00:03:47,459 --> 00:03:50,959
to the clean plate and the denoise plate,
67
00:03:50,959 --> 00:03:53,147
which gives us our difference.
68
00:03:53,147 --> 00:03:56,056
So it pushes that into the alpha channel.
69
00:03:56,056 --> 00:03:56,419
Again,
70
00:03:56,419 --> 00:04:00,783
that alpha is not very clean here so let's take a look
71
00:04:00,783 --> 00:04:02,056
at modifying that one more time.
72
00:04:02,056 --> 00:04:03,147
So here is the gray tool.
73
00:04:03,147 --> 00:04:09,146
Same thing, clamp the white alpha channel only, increase that gain.
74
00:04:09,147 --> 00:04:11,869
I'm going to aggressively push this out,
75
00:04:11,869 --> 00:04:16,146
blur it, and we're going to take a look at why.
76
00:04:16,147 --> 00:04:21,385
So if you're familiar with the preprocessing workflow you are going
77
00:04:21,385 --> 00:04:26,147
to shuffle the alpha into the rgba and invert it.
78
00:04:26,147 --> 00:04:27,813
So before we move forward,
79
00:04:27,813 --> 00:04:32,146
let's take a look at an alternative or an additional step that we
80
00:04:32,146 --> 00:04:36,296
have to modify before we add this matte into place.
81
00:04:36,296 --> 00:04:41,696
So with the denoise plate we have sampled an average of the
82
00:04:41,696 --> 00:04:44,146
overall backing region with a constant.
83
00:04:44,146 --> 00:04:46,146
We've taken a minus operator.
84
00:04:46,147 --> 00:04:50,196
We've compared the original denoise plate to the constant,
85
00:04:50,196 --> 00:04:55,146
which allows us to see where this backing region is uneven.
86
00:04:55,147 --> 00:04:59,146
So we can see that it is very uneven here.
87
00:04:59,146 --> 00:05:03,146
Wherever the image is too dark relative to this constant,
88
00:05:03,147 --> 00:05:06,147
it will make it brighter and wherever it's too light
89
00:05:06,147 --> 00:05:07,647
relative to the constant color,
90
00:05:07,647 --> 00:05:13,147
it will make it darker and the result will be an even backing region.
91
00:05:13,147 --> 00:05:17,480
So now we're going to take this image and we want to make
92
00:05:17,480 --> 00:05:21,147
sure that nothing in here accidentally will change our image
93
00:05:21,147 --> 00:05:24,147
that we're looking to really key.
94
00:05:24,147 --> 00:05:30,147
So we have taken our matte and multiplied it against this image,
95
00:05:30,147 --> 00:05:35,146
which basically prevents us from accidentally changing our image.
96
00:05:35,147 --> 00:05:37,369
Now you're going to notice again that it's very soft
97
00:05:37,369 --> 00:05:39,147
and we'll see why in just a moment.
98
00:05:39,147 --> 00:05:44,346
You're going to take this result and plus it against the denoise plate,
99
00:05:44,346 --> 00:05:47,146
which gives us the cleaner backing region.
100
00:05:47,147 --> 00:05:52,146
So just to see again, here's the original plate and here's the evened out plate.
101
00:05:52,147 --> 00:05:57,147
Now unfortunately, there are small changes we can see here.
102
00:05:57,147 --> 00:05:59,617
Maybe it's actually very hard to see,
103
00:05:59,617 --> 00:06:03,147
but believe me when I say that it's slightly off-color.
104
00:06:03,147 --> 00:06:08,146
So let's check to see if it shows up in one of those channels, RGB.
105
00:06:08,147 --> 00:06:12,480
It's a little bit in the blue channel, you can see those changes,
106
00:06:12,480 --> 00:06:17,396
and the reason why I softened and dilated out was for this reason.
107
00:06:17,396 --> 00:06:20,996
Let's take a look at that full color.
108
00:06:20,996 --> 00:06:24,146
Let's take a look at those details.
109
00:06:24,147 --> 00:06:28,346
So if I did not dilate out and I did not blur that initial alpha,
110
00:06:28,346 --> 00:06:31,620
look at how much detail that we lose in our edges
111
00:06:31,620 --> 00:06:35,410
before we even begin to pull a key.
112
00:06:35,410 --> 00:06:40,852
So definitely something to take into consideration and that just helps us,
113
00:06:40,852 --> 00:06:46,147
again even up the backing region and it allows us to pull that final matte.
114
00:06:46,147 --> 00:06:49,783
So now what we're going to do is now that we have that clean backing region,
115
00:06:49,783 --> 00:06:52,993
we're going to pull that luma key, so again,
116
00:06:52,993 --> 00:06:59,147
add that keyer, attach it to our cleaned up image.
117
00:06:59,147 --> 00:07:02,646
We are going to scroll down here so I'm going to
118
00:07:02,646 --> 00:07:04,874
focus on the top screen left corner,
119
00:07:04,874 --> 00:07:08,480
sample this corner over here so we're going to say 0.
120
00:07:08,480 --> 00:07:13,146
92 and I'm going to look at some of the smaller pixels over here,
121
00:07:13,146 --> 00:07:15,252
so I'm going to say 0.
122
00:07:15,252 --> 00:07:16,725
21 and enter that.
123
00:07:16,725 --> 00:07:18,568
Oh, we did that wrong.
124
00:07:18,568 --> 00:07:21,146
Let's erase 11, not a huge deal.
125
00:07:21,147 --> 00:07:23,146
Take a look at our alpha.
126
00:07:23,146 --> 00:07:27,146
So at first glance it's much cleaner than what we had previously.
127
00:07:27,147 --> 00:07:32,146
Let's just take a look at a finessed one, so here's one that I've modified.
128
00:07:32,147 --> 00:07:33,813
I have changed that alpha.
129
00:07:33,813 --> 00:07:38,480
So here's the one that we had and I've just brought this down so
130
00:07:38,480 --> 00:07:42,419
that those values are give or take lower than 0.
131
00:07:42,419 --> 00:07:42,828
05,
132
00:07:42,828 --> 00:07:48,147
so here's an optimized version and let's take a look at what that
133
00:07:48,147 --> 00:07:52,147
looks like in comparison to our original image.
134
00:07:52,147 --> 00:07:58,980
So here is our evened out backing region and a single matte versus here's
135
00:07:58,980 --> 00:08:03,147
our original one where the backing region was not cleaned.
136
00:08:03,147 --> 00:08:05,581
So let's take a look at our RGB.
137
00:08:05,581 --> 00:08:09,842
So here's our RGB and let's take a look at the top left-hand corner
138
00:08:09,842 --> 00:08:12,813
because this is really what we are worried about.
139
00:08:12,813 --> 00:08:17,146
So this is my refined backing region and this is my original one
140
00:08:17,147 --> 00:08:21,908
with all the holes in it or the uneven lighting.
141
00:08:21,908 --> 00:08:28,289
So already the evening out of the backing region is really doing a great job.
142
00:08:28,289 --> 00:08:31,146
So I'm going to take a minute and skip forward.
143
00:08:31,147 --> 00:08:34,147
We're going to pull a separate key for here,
144
00:08:34,147 --> 00:08:35,813
a separate key over here,
145
00:08:35,813 --> 00:08:40,070
and another one for the bottom down here and we're going to
146
00:08:40,070 --> 00:08:43,146
combine them all together with a keymix tool.
147
00:08:43,147 --> 00:08:45,771
So let's take a look at those over there.
148
00:08:45,771 --> 00:08:49,855
So if I scroll to the far screen right of my image and we're
149
00:08:49,855 --> 00:08:53,419
going to take a look at that finalized alpha.
150
00:08:53,419 --> 00:08:58,328
So here we have, you can see it's a little bit brighter,
151
00:08:58,328 --> 00:09:02,799
but again it's keeping to that rule of below 5% down here as well,
152
00:09:02,799 --> 00:09:06,147
below, oh, that looks like it's pushed a little bit far.
153
00:09:06,147 --> 00:09:10,060
Let's see what we can do to modify that.
154
00:09:10,060 --> 00:09:16,146
So we'll go into our alpha and I'll pull this in just a tad.
155
00:09:16,147 --> 00:09:17,946
There we go, and around there.
156
00:09:17,946 --> 00:09:22,146
So it's looking a lot better already and let's just look at that again,
157
00:09:22,147 --> 00:09:24,056
what that was looking like before.
158
00:09:24,056 --> 00:09:27,237
So here is before we added that last piece and
159
00:09:27,237 --> 00:09:29,147
after and look how that's growing.
160
00:09:29,147 --> 00:09:35,147
It's adding lots of nice detail in the bottom of our image.
161
00:09:35,147 --> 00:09:39,959
So now let's take a look at some refinements and our
162
00:09:39,959 --> 00:09:42,147
original versus our final matte.
163
00:09:42,147 --> 00:09:43,575
Now before we do that,
164
00:09:43,575 --> 00:09:48,147
I do want to show that I garbage matted out all of the excess backing region.
165
00:09:48,147 --> 00:09:54,146
Again, I used that handy gray tool so there is my final matte.
166
00:09:54,147 --> 00:09:56,527
I took the gray tool and did the same thing.
167
00:09:56,527 --> 00:09:59,146
No need to clamp the white because we're not affecting them,
168
00:09:59,147 --> 00:10:02,647
in fact we are actually trying to up the contrast.
169
00:10:02,647 --> 00:10:04,747
We are dilating out that matte,
170
00:10:04,747 --> 00:10:10,147
blurring it ever so slightly and using it as a whole dot matte for our image.
171
00:10:10,147 --> 00:10:14,147
Now if you want to soften that falloff, use the Gaussian.
172
00:10:14,147 --> 00:10:17,446
Just make sure that we're not losing too much detail in
173
00:10:17,446 --> 00:10:20,146
here so that's why I stuck to the box.
174
00:10:20,147 --> 00:10:26,747
Now let's take a look at the original that we had so here is our
175
00:10:26,747 --> 00:10:31,147
final image with our key and here is the original.
176
00:10:31,147 --> 00:10:34,608
Now take a look at the difference down here.
177
00:10:34,608 --> 00:10:36,146
Again, original, final matte.
178
00:10:36,147 --> 00:10:41,718
So it really pays off to take that extra time out to combine
179
00:10:41,718 --> 00:10:45,147
multiple mattes and even out the backing region.
180
00:10:45,147 --> 00:10:52,799
So one last test that's going to be the RGB so RGB again of the original,
181
00:10:52,799 --> 00:10:57,747
oh sorry, the final one versus the original so small details,
182
00:10:57,747 --> 00:11:02,146
especially in those edge works or in that edge work rather.
183
00:11:02,147 --> 00:11:10,146
And let's take a look at our final matte versus our image, our original image.
184
00:11:10,147 --> 00:11:12,371
So some loss of detail.
185
00:11:12,371 --> 00:11:13,706
Unfortunately that happens.
186
00:11:13,706 --> 00:11:18,155
We did say that the luma key doesn't necessarily do
187
00:11:18,155 --> 00:11:20,190
the best job with edge detail,
188
00:11:20,190 --> 00:11:25,600
but it gives us a really great start and a great place to work with and again
189
00:11:25,600 --> 00:11:34,600
combining those multiple mattes really makes a big difference.
16968
Can't find what you're looking for?
Get subtitles in any language from opensubtitles.com, and translate them here.