WEBVTT
Kind: captions
Language: en

1
00:00:00.000 --> 00:00:02.620
The dark doesn't look very good these days.

2
00:00:02.630 --> 00:00:07.129
And I'm really sorry if you're one of the
lucky ones who's never noticed this,

3
00:00:07.129 --> 00:00:09.130
because now I point it out,

4
00:00:09.130 --> 00:00:12.880
now I artificially increase the brightness
on the dark parts in this video

5
00:00:12.880 --> 00:00:17.120
so you can see what the shadows behind me
actually look like...

6
00:00:17.120 --> 00:00:19.460
now you'll notice.

7
00:00:19.470 --> 00:00:20.500
In movie trailers,

8
00:00:20.500 --> 00:00:22.950
in dark scenes in even prestige television,

9
00:00:22.950 --> 00:00:27.830
in YouTube videos where there are
nice, calm gradient backgrounds:

10
00:00:27.830 --> 00:00:31.240
you see this dreadful colour banding.

11
00:00:31.240 --> 00:00:32.700
There are three reasons for it.

12
00:00:32.700 --> 00:00:35.800
And the first is that
there aren't enough colours.

13
00:00:37.160 --> 00:00:39.540
...all right, I'll turn the lights on.

14
00:00:42.700 --> 00:00:44.200
In most modern digital video,

15
00:00:44.220 --> 00:00:49.040
there is a grand total of about
16.7 million possible colours.

16
00:00:49.040 --> 00:00:51.500
And that number comes from how your screen works.

17
00:00:51.500 --> 00:00:54.000
When you watch a video,
your phone, computer or TV

18
00:00:54.000 --> 00:00:56.860
takes that compressed digital signal
that's being sent to you,

19
00:00:56.860 --> 00:00:58.520
and it converts it into instructions.

20
00:00:58.520 --> 00:01:00.129
Those instructions go to the screen,

21
00:01:00.129 --> 00:01:02.450
which changes the brightness
of each of the millions of

22
00:01:02.450 --> 00:01:04.559
red, green and blue lights
that make up your screen,

23
00:01:04.559 --> 00:01:07.680
up to 60 times a second, perfectly.

24
00:01:07.680 --> 00:01:10.740
It's incredible technology that
we just take for granted.

25
00:01:10.740 --> 00:01:14.590
But those little lights can't be
adjusted to any brightness.

26
00:01:14.590 --> 00:01:16.540
The simplest digital signal
would just tell the screen

27
00:01:16.540 --> 00:01:18.860
whether to turn each light on or off.

28
00:01:18.860 --> 00:01:21.620
Each instruction would take a single bit,
a single one or zero.

29
00:01:21.620 --> 00:01:24.450
So for each pixel,
each combination of red green and blue,

30
00:01:24.450 --> 00:01:26.720
that gives you eight possible colours:

31
00:01:26.720 --> 00:01:28.610
two times two times two.

32
00:01:28.610 --> 00:01:30.330
Which doesn't look great.

33
00:01:30.330 --> 00:01:33.189
So, okay, let's add another bit for each colour.

34
00:01:33.189 --> 00:01:37.650
Two bits gives us four options for each light:
off, dark, sorta-bright, and completely on.

35
00:01:37.650 --> 00:01:41.360
Now we've got four times four times four,
64 colours.

36
00:01:41.360 --> 00:01:42.560
Still doesn't look great.

37
00:01:42.560 --> 00:01:45.750
For a modern screen,
you need eight bits for each pixel,

38
00:01:45.750 --> 00:01:48.689
so you've got 256 shades each
of red, green, and blue,

39
00:01:48.689 --> 00:01:51.619
which gives you those 16 million
colours that you're used to.

40
00:01:51.619 --> 00:01:53.330
That ought to be enough, right?

41
00:01:53.330 --> 00:01:55.110
It seemed to be, when the standards were written.

42
00:01:55.110 --> 00:01:56.110
And to be fair,

43
00:01:56.110 --> 00:02:00.160
every bit you add means more data to transmit
and more expensive equipment.

44
00:02:00.160 --> 00:02:02.490
It seemed like a pretty good compromise
at the time.

45
00:02:02.490 --> 00:02:04.869
Professionals, like people working
in digital cinema,

46
00:02:04.869 --> 00:02:07.229
they go further,
they might use ten bits for each pixel,

47
00:02:07.229 --> 00:02:10.429
giving them about a billion
total possible colours instead.

48
00:02:10.429 --> 00:02:14.819
And you may well have a fancy screen that
can show those more precise billion colours,

49
00:02:14.819 --> 00:02:18.959
but the picture quality is determined by the
weakest link in the chain.

50
00:02:18.959 --> 00:02:22.599
A ludicrously expensive HDR monitor isn't
going to fix the colour banding

51
00:02:22.599 --> 00:02:24.890
on that regular YouTube video.

52
00:02:24.890 --> 00:02:28.370
Anyway, it was decided that 256 shades
each of red, green, and blue,

53
00:02:28.370 --> 00:02:30.909
combined however you like,
that should be enough for most people.

54
00:02:30.909 --> 00:02:33.469
And for almost everything, it is.

55
00:02:33.469 --> 00:02:37.760
But at full resolution, this video is 1,920
pixels wide.

56
00:02:37.760 --> 00:02:40.499
And there are only 256 shades of green.

57
00:02:40.499 --> 00:02:42.560
So if I have a green gradient across
the whole video,

58
00:02:42.560 --> 00:02:46.349
even if you use every single shade of pure
green that's possible on this format,

59
00:02:46.349 --> 00:02:49.790
then you're still going to have a colour band
every seven or eight pixels.

60
00:02:49.790 --> 00:02:52.510
And if everything's much darker than that..

61
00:02:52.510 --> 00:02:55.340
suddenly you only might have half a dozen
different shades of green

62
00:02:55.340 --> 00:02:57.160
available across the whole picture.

63
00:02:57.160 --> 00:03:00.000
Even if you’re using a bit of red and blue
in there as well,

64
00:03:00.010 --> 00:03:03.209
there just aren’t that many
available dark colours.

65
00:03:03.209 --> 00:03:06.519
But if you've got 
plain, bright background,

66
00:03:06.519 --> 00:03:08.450
why doesn't that have
colour-banding all over it?

67
00:03:08.450 --> 00:03:11.749
Well, the second reason is
how human eyesight works.

68
00:03:11.749 --> 00:03:15.639
There is colour banding all over this video.
Right now.

69
00:03:15.639 --> 00:03:20.450
If we take a sample of the bright gradient
behind me and massively increase the contrast,

70
00:03:20.450 --> 00:03:23.199
you can see it's there…
but normally, it's invisible.

71
00:03:23.199 --> 00:03:27.099
And that’s because while the absolute difference
between two bands is still the same,

72
00:03:27.099 --> 00:03:28.999
the relative difference is tiny.

73
00:03:28.999 --> 00:03:33.249
It's the same reason that this chart looks
a lot closer than this chart.

74
00:03:33.249 --> 00:03:36.180
Going from 201 to 202
feels like a tiny change

75
00:03:36.180 --> 00:03:38.920
but going from 1 to 2 is a doubling.

76
00:03:38.920 --> 00:03:40.420
Even though in both cases,

77
00:03:40.430 --> 00:03:42.680
the absolute change is the same: one,

78
00:03:42.680 --> 00:03:46.260
we perceive one as being
bigger than the other.

79
00:03:46.260 --> 00:03:48.779
But even if you don't have enough colours,

80
00:03:48.779 --> 00:03:51.510
you can break up gradients another way,
you can use dithering.

81
00:03:51.510 --> 00:03:54.010
Instead of going from one colour to another
at a line,

82
00:03:54.010 --> 00:03:56.060
you can steadily transition from
one to the other,

83
00:03:56.060 --> 00:03:57.599
you can make the boundary fuzzy,

84
00:03:57.599 --> 00:04:00.809
so that from a distance
the colours appear to blend.

85
00:04:00.809 --> 00:04:03.909
And if you’re filming a real-life scene,
then that’ll happen in-camera,

86
00:04:03.909 --> 00:04:07.879
the natural light and the noise in the signal
and the sensor will do that for you.

87
00:04:07.879 --> 00:04:10.930
And it actually works really well,
it more or less solves the problem...

88
00:04:10.930 --> 00:04:13.569
but it won't look like that here.

89
00:04:13.569 --> 00:04:16.200
And that brings me to the third reason:
compression.

90
00:04:16.200 --> 00:04:18.900
A raw HD video, uncompressed,

91
00:04:18.900 --> 00:04:22.009
needs somewhere around
a gigabit per second of data.

92
00:04:22.009 --> 00:04:26.229
Even if you have an internet connection that
can somehow support streaming all that,

93
00:04:26.229 --> 00:04:29.419
it's incredibly wasteful and expensive to
use all that data.

94
00:04:29.419 --> 00:04:32.919
So every streaming service,
YouTube, Netflix, everyone,

95
00:04:32.919 --> 00:04:34.500
uses lossy compression:

96
00:04:34.500 --> 00:04:37.820
a program called an ‘encoder’ takes a
massive high-quality video file,

97
00:04:37.820 --> 00:04:41.210
and then throws away fine detail to save data.

98
00:04:41.210 --> 00:04:43.810
The more the video is compressed,
the worse it looks,

99
00:04:43.810 --> 00:04:45.949
but it’ll work on slower and slower connections,

100
00:04:45.949 --> 00:04:49.620
and it’ll be cheaper and cheaper
to run your streaming service.

101
00:04:49.620 --> 00:04:51.020
But at some point,

102
00:04:51.020 --> 00:04:54.870
the viewers are going to notice that the picture
doesn't look great.

103
00:04:54.870 --> 00:04:58.669
But the encoder, the compression software,
is written by very clever people:

104
00:04:58.669 --> 00:05:00.780
and it works out which bits of the scene

105
00:05:00.780 --> 00:05:02.591
the viewer is probably
going to be interested in,

106
00:05:02.591 --> 00:05:05.990
and puts most of its effort into making
that part look good.

107
00:05:05.990 --> 00:05:08.990
And that’s usually a bright, sharp part
of the scene,

108
00:05:08.990 --> 00:05:10.610
something well lit and in focus,

109
00:05:10.610 --> 00:05:12.919
and probably the bits that are moving.

110
00:05:12.919 --> 00:05:14.150
Now there are exceptions,

111
00:05:14.150 --> 00:05:17.520
where there's some slow, menacing threat
out of focus in the background,

112
00:05:17.520 --> 00:05:21.069
but for almost every scene,
if it’s dark, or out of focus,

113
00:05:21.069 --> 00:05:23.889
chances are the director and the viewer
doesn't care about it.

114
00:05:23.889 --> 00:05:25.620
So the encoder doesn’t care about it either:

115
00:05:25.620 --> 00:05:27.120
if there are only fifty kilobits,

116
00:05:27.120 --> 00:05:30.770
fifty thousand ones and zeros available to
describe each frame of video,

117
00:05:30.770 --> 00:05:33.260
then the encoder will make sure
that most of those

118
00:05:33.260 --> 00:05:37.080
are spent on where the viewer's
likely to be looking.

119
00:05:37.080 --> 00:05:39.979
All that dithering, all the fancy stuff
in the background?

120
00:05:39.979 --> 00:05:40.979
Too expensive.

121
00:05:40.980 --> 00:05:44.980
Smear it out into just big blocks
of solid colour, no-one'll notice.

122
00:05:44.980 --> 00:05:46.639
And we won't.

123
00:05:46.639 --> 00:05:48.930
Unless it's also very dark.

124
00:05:48.930 --> 00:05:50.629
At which point,

125
00:05:50.629 --> 00:05:54.540
there will be something nasty
lurking in the shadows.