Would you like to inspect the original subtitles? These are the user uploaded subtitles that are being translated:
1
00:00:00,512 --> 00:00:03,328
Now they're likely to be more control Laura's coming through
2
00:00:03,584 --> 00:00:04,608
In the next few weeks
3
00:00:05,632 --> 00:00:07,680
The original controllers control
4
00:00:07,936 --> 00:00:08,704
And control
5
00:00:08,960 --> 00:00:11,264
Can be used by Universe similar way in fact
6
00:00:11,776 --> 00:00:12,800
I used to control net
7
00:00:14,336 --> 00:00:18,432
At work flow that we use for the controller for the canny
8
00:00:18,688 --> 00:00:19,456
Control Laur
9
00:00:19,712 --> 00:00:21,504
And this was the result that I got
10
00:00:21,760 --> 00:00:22,528
Using the
11
00:00:23,296 --> 00:00:25,856
Best control net that has been interested for STX l
12
00:00:26,368 --> 00:00:28,416
This is in comparison
13
00:00:28,672 --> 00:00:32,000
Play control Laura the 128 Frank control Laura from
14
00:00:32,256 --> 00:00:33,280
Stability a
15
00:00:33,536 --> 00:00:35,328
So there's quite a lot of difference between the two
16
00:00:35,840 --> 00:00:36,864
And I would
17
00:00:37,120 --> 00:00:41,216
Play for this particular render I think the control Laura actually won a little bit
18
00:00:41,472 --> 00:00:44,032
The original image is this image here
19
00:00:44,288 --> 00:00:44,800
We are looking
20
00:00:46,080 --> 00:00:48,384
Put very nice 1024 / 1024 image
21
00:00:48,896 --> 00:00:51,200
And I think what we want to do with this one
22
00:00:51,456 --> 00:00:52,736
Is to take a look at
23
00:00:52,992 --> 00:00:55,040
How we can get better results Maybe
24
00:00:55,552 --> 00:00:56,320
Even then this
25
00:00:56,576 --> 00:00:58,112
Using the controller
26
00:00:58,368 --> 00:00:59,392
What are the control
27
00:00:59,904 --> 00:01:02,208
First introduced
28
00:01:02,464 --> 00:01:04,512
What's in February of this year
29
00:01:04,768 --> 00:01:06,048
And they were introduced
30
00:01:06,304 --> 00:01:09,120
What's the following research at Stanford University
31
00:01:09,376 --> 00:01:11,424
The research paper is one
32
00:01:12,192 --> 00:01:13,984
Is fairly interesting to look at
33
00:01:14,240 --> 00:01:16,288
And having done that I can show you
34
00:01:16,544 --> 00:01:18,592
The original concept so we
35
00:01:18,848 --> 00:01:20,384
What's the county Edge detection here
36
00:01:20,896 --> 00:01:24,992
Is one that we've looked at for the controllers and we're going to be looking at for the control Nets
37
00:01:25,504 --> 00:01:27,808
And you can see the basic idea
38
00:01:29,088 --> 00:01:29,600
Control
39
00:01:29,856 --> 00:01:31,136
Set an aspects of the
40
00:01:31,648 --> 00:01:33,696
Using the original image
41
00:01:34,208 --> 00:01:37,536
Now I want to take a look at other ways that you can use control Nets cuz
42
00:01:37,792 --> 00:01:40,352
Very well presented inside of this paper
43
00:01:40,864 --> 00:01:42,912
So there are more examples of the
44
00:01:43,168 --> 00:01:43,936
Play Connie
45
00:01:44,448 --> 00:01:45,472
And how it can be used to
46
00:01:45,728 --> 00:01:47,264
Can you generate different types of images
47
00:01:47,776 --> 00:01:48,800
And here we've got
48
00:01:49,824 --> 00:01:50,336
Results
49
00:01:50,848 --> 00:01:52,896
Working with automatic pumps and user
50
00:01:53,920 --> 00:01:58,272
And you can see it I really like this one and that one and you can see the cute kittens
51
00:01:58,784 --> 00:02:00,320
Some variations
52
00:02:00,576 --> 00:02:01,344
It's based on
53
00:02:01,856 --> 00:02:02,624
The
54
00:02:02,880 --> 00:02:05,952
What's the basic composition remaining the same
55
00:02:06,208 --> 00:02:09,280
So this is really about retaining compositional elements
56
00:02:09,536 --> 00:02:11,584
And we can see here also that you can
57
00:02:11,840 --> 00:02:12,608
Use
58
00:02:13,120 --> 00:02:14,144
That have been
59
00:02:14,656 --> 00:02:18,240
That have been created inside of stability
60
00:02:18,496 --> 00:02:19,520
That's stable diffusion
61
00:02:20,032 --> 00:02:23,360
What are the AI images to create the socials
62
00:02:26,688 --> 00:02:29,248
But the paper doesn't stop there it shows us how
63
00:02:29,504 --> 00:02:30,528
Line input
64
00:02:30,784 --> 00:02:32,576
Can be used to create variations
65
00:02:33,088 --> 00:02:34,880
And for the down we can see
66
00:02:35,392 --> 00:02:36,160
Other uses
67
00:02:36,928 --> 00:02:38,208
Basic scribbles
68
00:02:38,720 --> 00:02:39,232
How to
69
00:02:39,488 --> 00:02:40,768
Play Adele's variations
70
00:02:42,048 --> 00:02:46,144
The amount of research that went into this very simple elephant
71
00:02:47,680 --> 00:02:48,704
Produces these
72
00:02:49,216 --> 00:02:50,240
These variations here
73
00:02:50,496 --> 00:02:54,848
Has been going on for quite some time and quite a lot of what we're doing
74
00:02:55,360 --> 00:02:56,384
Basically
75
00:02:56,640 --> 00:02:58,176
Is built on this type of research
76
00:02:58,688 --> 00:03:03,552
But it's interesting to see how you can sometimes get less and less of research
77
00:03:04,064 --> 00:03:06,624
Kind of coming together in a single project
78
00:03:07,392 --> 00:03:09,696
Which certainly what seems to have happened here
79
00:03:09,952 --> 00:03:13,280
So there's a type of input known as the Hedge
80
00:03:15,328 --> 00:03:16,608
Another one is
81
00:03:16,864 --> 00:03:17,376
Play pose
82
00:03:18,912 --> 00:03:20,192
These women posing
83
00:03:20,704 --> 00:03:22,240
According to this kind of stick figure
84
00:03:23,264 --> 00:03:24,800
Add more posing
85
00:03:25,568 --> 00:03:26,592
So all of this is
86
00:03:26,848 --> 00:03:29,920
What's possible with the control net
87
00:03:30,432 --> 00:03:31,968
The posing head is
88
00:03:32,224 --> 00:03:33,760
Who's coming from a
89
00:03:35,808 --> 00:03:37,344
Michael Jackson's concert
90
00:03:39,392 --> 00:03:40,160
Turn lights to see
91
00:03:40,416 --> 00:03:41,696
The king strutting his
92
00:03:41,952 --> 00:03:42,976
Off in AI
93
00:03:43,488 --> 00:03:49,376
And we've also got some really basic stuff this kind of stuff I think this might be called segmentation
94
00:03:49,632 --> 00:03:50,656
You have to
95
00:03:51,680 --> 00:03:52,960
You'd have to
96
00:03:53,216 --> 00:03:55,264
Correct me on that if I'm wrong but you can see
97
00:03:55,520 --> 00:03:56,544
Play basic color
98
00:03:57,056 --> 00:03:58,336
I'm coming through
99
00:03:58,592 --> 00:03:59,872
Impressive images
100
00:04:00,128 --> 00:04:02,432
And what the research is the same
101
00:04:06,784 --> 00:04:09,344
It's kind of research has been done by several researchers
102
00:04:09,600 --> 00:04:10,880
Overtime Nvidia
103
00:04:11,392 --> 00:04:13,184
Stanford University
104
00:04:13,440 --> 00:04:15,744
And interestingly they compare
105
00:04:16,000 --> 00:04:20,095
Technique I think this is a technique which was developed by stability AI
106
00:04:20,863 --> 00:04:22,399
And we can see
107
00:04:22,655 --> 00:04:24,703
How did Technic actually compares
108
00:04:25,215 --> 00:04:26,751
The benefits and the
109
00:04:28,031 --> 00:04:29,055
And the disadvantages
110
00:04:29,311 --> 00:04:30,847
So quite a lot of stuff here
111
00:04:31,103 --> 00:04:33,151
If you're more research-oriented
112
00:04:33,407 --> 00:04:35,967
You probably want to take a look at this this paper
113
00:04:37,759 --> 00:04:39,551
What we want to do however is to look at
114
00:04:40,319 --> 00:04:41,599
Fatalities of using
115
00:04:42,367 --> 00:04:43,135
Control net
116
00:04:44,159 --> 00:04:48,255
One thing I wanted to mention with the controllers for some of the more advanced users out there
117
00:04:48,767 --> 00:04:51,071
Is that they do mention just how easy it is to
118
00:04:51,327 --> 00:04:52,607
Play train a control net so
119
00:04:53,887 --> 00:05:00,031
They say training is as fast as fine-tuning at the fusion model so if you want to get a little bit more sophisticated or advanced
120
00:05:00,287 --> 00:05:02,847
It should be possible to create your own control net
121
00:05:03,103 --> 00:05:05,151
But in future it may be possible ALS
122
00:05:06,175 --> 00:05:07,967
How to create your own control Laura
123
00:05:08,223 --> 00:05:10,783
Which can be derived from control net
124
00:05:11,551 --> 00:05:14,879
Came out a couple of weeks ago
125
00:05:15,135 --> 00:05:15,903
And if you want
126
00:05:16,415 --> 00:05:17,439
Download that one
127
00:05:17,695 --> 00:05:18,975
It is cold
128
00:05:19,231 --> 00:05:21,279
I think actually published by hugging face
129
00:05:21,535 --> 00:05:22,303
Go control
130
00:05:22,559 --> 00:05:24,351
Put candy sdx 0
131
00:05:24,607 --> 00:05:25,631
Dash 1.0
132
00:05:25,887 --> 00:05:29,983
So this one is the one you download I'll have a link to this particular page
133
00:05:30,495 --> 00:05:33,823
And there are a number of options they've got here so the one I will be using
134
00:05:34,079 --> 00:05:36,127
I think it's going to be the safe tensor
135
00:05:36,639 --> 00:05:39,711
5 gigabyte one away it might be the pickle 5 gigabyte
136
00:05:40,223 --> 00:05:41,247
One of those two
137
00:05:41,503 --> 00:05:45,599
And you can if you want to download the 2 1/2 GB versions of these on
138
00:05:47,391 --> 00:05:48,671
Heard some feedback
139
00:05:50,207 --> 00:05:56,351
Different uses not everyone is getting success with smaller ones so the larger ones I've not had
140
00:05:56,607 --> 00:05:58,143
Play anything but success with those logic
141
00:05:58,911 --> 00:06:02,495
The project itself contains quite a lot of information about how
142
00:06:03,007 --> 00:06:04,287
The control that can be used
143
00:06:04,799 --> 00:06:07,615
But we'll cover the basic use patterns in this video
144
00:06:08,127 --> 00:06:09,407
Oh and following videos
145
00:06:09,663 --> 00:06:11,455
And I'll give you some
146
00:06:11,711 --> 00:06:16,319
Files that you can use to set up your own control nest and get some pretty decent results
147
00:06:19,391 --> 00:06:22,463
Now conceptually controlling is a very similar to control Laura
148
00:06:22,719 --> 00:06:26,559
And if you haven't watched the videos on control Laura's I would recommend that you watch those
149
00:06:27,327 --> 00:06:29,119
Working with controllers
150
00:06:29,375 --> 00:06:30,143
Because everything
151
00:06:30,399 --> 00:06:31,935
We need to do for the control
152
00:06:32,191 --> 00:06:33,727
What's needs to be understood
153
00:06:35,007 --> 00:06:37,055
From the videos inside of the control
154
00:06:37,311 --> 00:06:39,103
What is the same process you need to download
155
00:06:39,359 --> 00:06:40,895
The same extensions
156
00:06:41,407 --> 00:06:42,175
And
157
00:06:42,431 --> 00:06:45,759
Will show you in the next video how to actually put everything together
10873
Can't find what you're looking for?
Get subtitles in any language from opensubtitles.com, and translate them here.