Would you like to inspect the original subtitles? These are the user uploaded subtitles that are being translated:
1
00:00:00,256 --> 00:00:04,352
An Innovative new control Laura from stability AI is the
2
00:00:04,864 --> 00:00:09,216
Revision control Laura and I can take an image like this here
3
00:00:09,728 --> 00:00:14,848
The one that you sing Hit and then create a number of images from it so here we're seeing
4
00:00:15,360 --> 00:00:18,176
Four Images which were created from that image
5
00:00:19,456 --> 00:00:25,600
Text prompt not the ones that you're looking at here come from a different inspiration and
6
00:00:25,856 --> 00:00:26,880
And we will take a look
7
00:00:27,136 --> 00:00:28,416
At that one shortly
8
00:00:30,976 --> 00:00:35,072
Silhouette of a man
9
00:00:35,584 --> 00:00:37,376
Leaving across to
10
00:00:39,680 --> 00:00:41,216
Tops of mountains or
11
00:00:45,568 --> 00:00:48,384
What I did was that I put in some
12
00:00:48,640 --> 00:00:50,432
Put on the default settings
13
00:00:50,944 --> 00:00:52,224
Let it run
14
00:00:52,480 --> 00:00:55,040
It came out basically with the ideas
15
00:00:55,296 --> 00:00:57,088
I think captured
16
00:00:57,344 --> 00:00:59,136
Silhouette
17
00:00:59,392 --> 00:01:02,464
And sunset so all the images that we're looking at
18
00:01:02,720 --> 00:01:04,768
I've got some degree of sunset going on
19
00:01:05,024 --> 00:01:06,304
And as well as the sun
20
00:01:06,560 --> 00:01:07,584
Feature going on
21
00:01:07,840 --> 00:01:10,912
We have a certain amount of mountains
22
00:01:11,168 --> 00:01:15,264
So I think it figured out mountains sunset
23
00:01:16,544 --> 00:01:17,312
Silhouette
24
00:01:17,568 --> 00:01:18,080
And I don't
25
00:01:18,336 --> 00:01:22,432
I think it really got the leaping this one where the gentleman is sleeping
26
00:01:22,688 --> 00:01:27,040
I'm actually asked for a lady and it was like yeah no you're going to have a guy
27
00:01:27,296 --> 00:01:28,832
That might be a lady I don't know
28
00:01:31,136 --> 00:01:35,744
Taking a look at the image trying to figure out what's in the image and then creating
29
00:01:36,256 --> 00:01:38,048
Revisions from that image
30
00:01:38,560 --> 00:01:43,936
Strange amount of texts going on as well so let's try again this time we'll put in
31
00:01:45,216 --> 00:01:45,984
Text
32
00:01:48,288 --> 00:01:49,312
Add Water
33
00:01:50,336 --> 00:01:54,944
And what will do as well will reduce the strength of the conditioning
34
00:02:01,856 --> 00:02:03,392
Maybe add a bit of noise as well
35
00:02:08,000 --> 00:02:13,632
We'll run that see what happens now this one uses a type of file which we haven't come across before
36
00:02:14,144 --> 00:02:16,192
Known as Eclipse vision
37
00:02:16,448 --> 00:02:17,984
File and this one
38
00:02:18,240 --> 00:02:22,592
Is the clip Vision G I'll show you where to download it in a moment
39
00:02:23,104 --> 00:02:24,896
And maybe
40
00:02:25,664 --> 00:02:29,248
We can take a quick look at how it all connects up together
41
00:02:29,504 --> 00:02:34,112
Some of these are very difficult to understand you sort of need to sit at the
42
00:02:34,368 --> 00:02:37,952
Screen and then start playing around with them it's really understand how everything is tied together
43
00:02:38,208 --> 00:02:41,024
But essentially we've got the main
44
00:02:41,536 --> 00:02:44,096
Elements here which is doing most of the work
45
00:02:46,144 --> 00:02:52,288
And because it's producing four Images let's take a look at the images here we can see it sort of gallery going
46
00:02:52,544 --> 00:02:53,056
Going on here
47
00:02:54,336 --> 00:02:56,384
You can just click away close it up
48
00:02:56,640 --> 00:03:02,016
And I think because we've got another Gallery installed we can also double click and use that gallery
49
00:03:02,528 --> 00:03:03,552
To see what's happening
50
00:03:03,808 --> 00:03:06,112
Now what will do let's get this running
51
00:03:06,368 --> 00:03:12,512
And you can see it it's not fast at all it's got to produce four Images both of
52
00:03:12,768 --> 00:03:14,560
All of them about the same size as the original
53
00:03:14,816 --> 00:03:17,376
So it will take a little bit longer than might be
54
00:03:17,632 --> 00:03:20,192
Assumed from the simple 30 steps
55
00:03:22,752 --> 00:03:24,544
Add the process of
56
00:03:24,800 --> 00:03:26,336
Actually getting everything in place
57
00:03:27,104 --> 00:03:29,408
So instability AI we've got
58
00:03:29,664 --> 00:03:31,456
Some information about how everything
59
00:03:31,712 --> 00:03:34,016
Moves together how everything works together
60
00:03:35,808 --> 00:03:39,392
First of all they say revision this is the revision section
61
00:03:39,904 --> 00:03:44,000
Revision is a novel approach of using
62
00:03:44,512 --> 00:03:47,072
Images to images to prompt sdx
63
00:03:47,328 --> 00:03:50,144
Its uses pulled clip embeddings
64
00:03:50,656 --> 00:03:53,984
To produce images conceptually similar to the input
65
00:03:54,240 --> 00:03:58,080
It can be used either in addition or to replace text prompts
66
00:03:58,336 --> 00:04:03,456
Didn't have much of an effect because I was asking for a female
67
00:04:03,712 --> 00:04:09,600
And I was saying look with you we don't want to mail so it did not overwrite that particular
68
00:04:11,392 --> 00:04:14,720
That particular observation that we had a male in the
69
00:04:18,303 --> 00:04:21,887
Now let's take a look at what you need to actually download in order to
70
00:04:22,143 --> 00:04:24,447
Get this setup is not quite as straightforward
71
00:04:24,703 --> 00:04:25,727
As the other ones
72
00:04:26,239 --> 00:04:28,799
If we go and take a look at revision
73
00:04:29,311 --> 00:04:32,895
We can see we need to download this clip Vision G safe tenses
74
00:04:33,151 --> 00:04:35,711
And this one is huge is 3.69
75
00:04:35,967 --> 00:04:41,343
So that one is the one that we need to actually put into this section here
76
00:04:44,159 --> 00:04:49,535
Into this section here and there's a specific directory that we put it in inside of
77
00:04:49,791 --> 00:04:50,815
I'll come for you I
78
00:04:52,863 --> 00:04:55,935
Now in the folder structure navigate to
79
00:04:56,191 --> 00:04:59,519
Configure I go to models and go to clip vision
80
00:04:59,775 --> 00:05:02,335
That's why we dropped the eclipse Vision G file
81
00:05:02,847 --> 00:05:05,151
So that one should
82
00:05:05,407 --> 00:05:07,967
Think be about ready so let's take a look
83
00:05:13,343 --> 00:05:14,111
Okay it's finished
84
00:05:14,879 --> 00:05:15,647
That wasn't too bad
85
00:05:16,159 --> 00:05:18,719
And as you can see here we've got four images
86
00:05:18,975 --> 00:05:25,119
Let's take a look at the prompt that I put in a female leaping don't clouds Sunset Birds male text water
87
00:05:25,631 --> 00:05:26,911
As on negatives
88
00:05:27,423 --> 00:05:28,959
So we've certainly got
89
00:05:29,215 --> 00:05:33,055
A male here and there's someone confusing confusing shadow
90
00:05:33,567 --> 00:05:36,895
We've got the dawn happening clouds
91
00:05:37,151 --> 00:05:37,919
Definitely
92
00:05:42,271 --> 00:05:43,807
Who knows what's Happening Here
93
00:05:44,063 --> 00:05:45,343
Interesting
94
00:05:45,855 --> 00:05:47,647
And this one is beautiful but
95
00:05:47,903 --> 00:05:50,463
It doesn't have the key elements which is the
96
00:05:51,487 --> 00:05:52,767
The female leaping
97
00:05:53,535 --> 00:05:57,887
It doesn't have those key elements it looks beautiful what a little bit of text down here
98
00:06:00,447 --> 00:06:02,751
Then we've got a really nice looking image here
99
00:06:03,007 --> 00:06:06,079
So is it is interesting but I think
100
00:06:06,591 --> 00:06:08,895
It's a little bit Wild
101
00:06:12,223 --> 00:06:13,503
I think these ones were
102
00:06:13,759 --> 00:06:18,111
Exceptionally organized and you know
103
00:06:18,367 --> 00:06:20,415
Now compared to the original image they were
104
00:06:20,927 --> 00:06:23,231
Broadly speaking similar
105
00:06:23,743 --> 00:06:25,535
And yeah we can see
106
00:06:29,375 --> 00:06:35,519
I had a bit of noise it it is interesting it's an interesting
107
00:06:35,775 --> 00:06:38,847
I think this one is going to require quite a bit of experimentation
108
00:06:40,639 --> 00:06:41,407
How
109
00:06:42,687 --> 00:06:47,807
How it works and I can understand that the putting in batch size of four
110
00:06:48,063 --> 00:06:48,831
For this one
111
00:06:49,087 --> 00:06:50,879
I think for
112
00:06:51,903 --> 00:06:56,255
You may want to reduce the back size down to one
113
00:06:56,511 --> 00:06:59,071
Best size 4 is just a default
114
00:06:59,583 --> 00:07:01,631
And what we can do
115
00:07:01,887 --> 00:07:05,215
Is then run the prom several times perhaps and then we'll get
116
00:07:05,471 --> 00:07:09,567
We'll get one image each time but you might not be quite as much of a
117
00:07:10,591 --> 00:07:12,127
For your graphics card
118
00:07:12,383 --> 00:07:17,247
So I'll share this one with you probably revert this
119
00:07:26,719 --> 00:07:27,487
210
120
00:07:27,743 --> 00:07:32,351
And I'll share this with you so that you can test it with your own with your own
121
00:07:32,607 --> 00:07:38,751
Images now there's another use case for this one and it's a little bit more complicated will take
122
00:07:39,007 --> 00:07:39,775
Go look at this one
123
00:07:40,031 --> 00:07:40,543
Next
9463
Can't find what you're looking for?
Get subtitles in any language from opensubtitles.com, and translate them here.