Would you like to inspect the original subtitles? These are the user uploaded subtitles that are being translated:
1
00:00:00,256 --> 00:00:04,096
A non-standard workflow
2
00:00:05,120 --> 00:00:07,168
Sdx will using configure
3
00:00:07,424 --> 00:00:12,032
And the two refined the two models the base model and the refiner model
4
00:00:12,288 --> 00:00:17,152
Was one of the early workflows was adapted actually from one of the early workflows
5
00:00:18,176 --> 00:00:19,200
What
6
00:00:20,480 --> 00:00:24,576
Someone created this I can't remember exactly who around the time that the
7
00:00:25,344 --> 00:00:27,648
Sdx beta version came out
8
00:00:27,904 --> 00:00:31,488
And I decided to play around with it I created a wide
9
00:00:31,744 --> 00:00:33,280
Which version of
10
00:00:33,536 --> 00:00:34,304
Landscape
11
00:00:35,328 --> 00:00:41,472
And once was supposed to really be working with images that are around 1,024 by 1,024
12
00:00:43,264 --> 00:00:44,288
1 megapixel
13
00:00:45,568 --> 00:00:46,336
This
14
00:00:46,592 --> 00:00:47,616
Extend things quite
15
00:00:48,128 --> 00:00:51,456
Extends to 2 megapixels roughly 2 megapixels
16
00:00:52,224 --> 00:00:54,272
With a landscape design
17
00:00:54,784 --> 00:00:56,320
Nearly two to one landscape
18
00:00:56,576 --> 00:00:57,600
Now with this one
19
00:00:57,856 --> 00:00:58,368
The
20
00:00:58,624 --> 00:00:59,136
What comes
21
00:00:59,392 --> 00:00:59,904
Pretty good
22
00:01:02,720 --> 00:01:04,000
Randomness of the
23
00:01:05,280 --> 00:01:06,304
Hear the prompt is
24
00:01:06,560 --> 00:01:09,120
One which includes a lot of detail
25
00:01:09,888 --> 00:01:13,472
Announce the software to really kind of play around with
26
00:01:13,728 --> 00:01:15,008
The ingredients
27
00:01:15,264 --> 00:01:16,288
To come up with something
28
00:01:16,800 --> 00:01:18,080
Usually involves a woman
29
00:01:18,336 --> 00:01:21,152
And a mirror and a lot of plants and butterflies
30
00:01:22,176 --> 00:01:22,944
Even though
31
00:01:23,200 --> 00:01:24,480
Typically what happens when you
32
00:01:24,736 --> 00:01:27,552
Can you increase the size outside of that sort of 1 m p
33
00:01:29,856 --> 00:01:31,392
Is clever prompt you can actually
34
00:01:31,648 --> 00:01:32,416
Keep get things
35
00:01:34,464 --> 00:01:35,488
It doesn't matter
36
00:01:35,744 --> 00:01:36,768
Weather
37
00:01:37,024 --> 00:01:38,304
What's the prompting and the
38
00:01:38,560 --> 00:01:39,328
Design
39
00:01:39,840 --> 00:01:43,168
Follow the older dogs method of around 1 megapixel
40
00:01:43,680 --> 00:01:44,448
If it works it
41
00:01:44,960 --> 00:01:48,032
Now with this one there are some issues with the outcome
42
00:01:48,288 --> 00:01:49,568
For instance with
43
00:01:49,824 --> 00:01:51,872
At what time are you working at the moment on
44
00:01:52,896 --> 00:01:54,432
Did this particular one which is rendering
45
00:01:54,944 --> 00:01:58,016
And that will produce a refined version here
46
00:01:58,272 --> 00:01:59,040
But this is
47
00:01:59,296 --> 00:02:01,088
Play previous version which was refined
48
00:02:01,600 --> 00:02:03,392
You can see it's kind of completely
49
00:02:05,184 --> 00:02:05,952
The designs
50
00:02:06,208 --> 00:02:07,488
Completely filled everything
51
00:02:08,000 --> 00:02:10,304
In in that 2 megapixel image
52
00:02:10,560 --> 00:02:12,096
Which approximately 2 megapixel image
53
00:02:13,632 --> 00:02:16,960
We've got the mirror we've got lots of Ivy we've got lots of butterflies
54
00:02:17,472 --> 00:02:19,776
As long as things just make sense
55
00:02:21,312 --> 00:02:25,920
And you can cram things into a prompt sometimes you can go beyond that sort of 1 megapixel
56
00:02:26,944 --> 00:02:27,712
A limit
57
00:02:27,968 --> 00:02:30,528
And you can actually get some pretty decent outcomes
58
00:02:31,040 --> 00:02:35,392
This one is pretty good I think one area where we had a little bit of a problem was
59
00:02:35,648 --> 00:02:37,696
I think the woman the woman looks good
60
00:02:42,816 --> 00:02:44,608
The women
61
00:02:44,864 --> 00:02:46,912
Often have long necks I'm not sure why
62
00:02:47,424 --> 00:02:48,704
And the prompt
63
00:02:48,960 --> 00:02:50,496
How to prompt actually contains long neck
64
00:02:51,264 --> 00:02:53,056
Another thing that it did was to
65
00:02:53,568 --> 00:02:56,896
Stop with the hand not really
66
00:02:59,712 --> 00:03:00,736
What's happening with the
67
00:03:00,992 --> 00:03:02,528
So she's behind the mirror
68
00:03:02,784 --> 00:03:03,296
She's supposed to
69
00:03:03,552 --> 00:03:04,832
What kind of coming out of the mirror
70
00:03:05,344 --> 00:03:06,368
And it's sort of
71
00:03:06,624 --> 00:03:07,392
The render
72
00:03:07,648 --> 00:03:09,696
Add step 28
73
00:03:10,464 --> 00:03:13,536
And I think actually this is the the
74
00:03:13,792 --> 00:03:16,608
Refined version is stopped rendering it to step 23
75
00:03:17,120 --> 00:03:19,680
And the Hand wasn't quite well formed
76
00:03:20,192 --> 00:03:25,568
In this type of situation what you might want to do is to increase the number of steps because I can sometimes give it an opportunity
77
00:03:26,080 --> 00:03:26,848
Talk to fix any problems
78
00:03:27,104 --> 00:03:30,176
There are other ways of dealing with problems like this for instance you
79
00:03:30,688 --> 00:03:32,992
This one you could easily going to photoshop and just
80
00:03:33,504 --> 00:03:35,552
Copy some of the Ivy and just cover that
81
00:03:35,808 --> 00:03:36,832
That wouldn't be a problem
82
00:03:37,600 --> 00:03:39,392
The other thing you could do we can use
83
00:03:40,416 --> 00:03:42,720
A type of model or type of
84
00:03:43,488 --> 00:03:44,768
Scheme known as an em
85
00:03:45,024 --> 00:03:46,048
Which can sometimes
86
00:03:46,816 --> 00:03:47,328
What tack
87
00:03:47,840 --> 00:03:49,120
Difficult issues
88
00:03:50,912 --> 00:03:51,680
For instance
89
00:03:53,984 --> 00:03:55,264
Bad looking hands
90
00:03:57,568 --> 00:03:58,848
And in fact I have put
91
00:03:59,104 --> 00:04:00,128
The prompts here
92
00:04:00,384 --> 00:04:01,408
Sometimes work
93
00:04:01,920 --> 00:04:03,968
And I find that the prompt
94
00:04:04,480 --> 00:04:07,040
What disgusting in the negative prompt
95
00:04:07,296 --> 00:04:08,064
Area
96
00:04:09,600 --> 00:04:10,112
Weird
97
00:04:10,368 --> 00:04:15,488
Those are sometimes very very helpful prompts because they just seem to tell the
98
00:04:17,791 --> 00:04:19,327
Don't don't have anything that looks
99
00:04:19,583 --> 00:04:20,095
Looks weird
100
00:04:20,351 --> 00:04:21,375
Anything that looks strange
101
00:04:21,631 --> 00:04:24,191
And it seems sometimes to solve problems
102
00:04:24,447 --> 00:04:29,311
Now in this particular design I'm trying to work with something that's fairly surrealistic
103
00:04:29,823 --> 00:04:30,591
So I don't want
104
00:04:30,847 --> 00:04:31,871
What's the prompts
105
00:04:32,383 --> 00:04:33,919
And then the negative prom
106
00:04:38,015 --> 00:04:39,551
Emit anything
107
00:04:39,807 --> 00:04:43,135
That might be surrealistic so weird might not be a negative
108
00:04:43,647 --> 00:04:45,695
Prompt that I really should have there because
109
00:04:48,511 --> 00:04:49,279
In a port
110
00:04:51,327 --> 00:04:52,607
Coming out of a window
111
00:04:52,863 --> 00:04:53,631
Coming out of a mere
112
00:04:54,143 --> 00:04:56,703
It's weird and you don't want to restrict the
113
00:04:59,007 --> 00:04:59,775
The creative
114
00:05:00,287 --> 00:05:02,591
A element within the the software
115
00:05:03,103 --> 00:05:04,383
The prompting needs to be careful
116
00:05:04,639 --> 00:05:05,663
Now this particular
117
00:05:07,967 --> 00:05:10,527
Is one which uses a very unusual
118
00:05:11,807 --> 00:05:15,135
It uses an unusual unusual process
119
00:05:15,391 --> 00:05:16,159
We start off
120
00:05:16,671 --> 00:05:18,463
A case sampler for the
121
00:05:18,975 --> 00:05:22,559
Base model and then we have another case sampler for the refined model
122
00:05:23,071 --> 00:05:24,607
So it's actually taking
123
00:05:25,887 --> 00:05:27,935
Base model and then it's sort of like putting it
124
00:05:28,447 --> 00:05:31,007
Into a similar process for the refined model
125
00:05:31,263 --> 00:05:32,543
This is non-standard
126
00:05:33,055 --> 00:05:33,567
But
127
00:05:33,823 --> 00:05:36,127
Wanted to do was to give you this particular
128
00:05:39,455 --> 00:05:42,783
Even though it's not really how we should be using
129
00:05:43,039 --> 00:05:44,319
St XL
130
00:05:44,575 --> 00:05:48,415
The theory behind it is that you use the
131
00:05:50,975 --> 00:05:55,071
Models in a very specific way with a base model doing most of the work
132
00:05:55,327 --> 00:05:57,631
And then the refiner model doing a tiny amount of work
133
00:05:58,143 --> 00:06:00,703
This is a an unusual setup which
134
00:06:02,239 --> 00:06:05,823
I found produce reasonably good results most of the time
135
00:06:06,079 --> 00:06:11,455
The entire process is taking a long time and that's because I've got my software for recording
136
00:06:12,479 --> 00:06:13,759
Running in the background
137
00:06:14,015 --> 00:06:15,039
And also
138
00:06:15,295 --> 00:06:18,367
This is a very long image if you're working with something like
139
00:06:18,623 --> 00:06:19,647
Like a 8 GB
140
00:06:20,159 --> 00:06:23,999
Graphics cards graphics card with 8 GB of RAM
141
00:06:25,023 --> 00:06:26,815
Is it going to start struggling
142
00:06:27,071 --> 00:06:28,607
Beyond that one megapixel
143
00:06:29,887 --> 00:06:33,727
So what happens in that situation is that you will get a bottleneck and the bottleneck
144
00:06:34,495 --> 00:06:36,031
Can actually occur with a case samp
145
00:06:36,287 --> 00:06:37,311
Most of the time
146
00:06:37,567 --> 00:06:38,847
Did the bottom like a curse with
147
00:06:39,103 --> 00:06:40,383
Vae decoder
148
00:06:42,431 --> 00:06:43,199
Variational
149
00:06:44,223 --> 00:06:45,503
Encoder decoder
150
00:06:46,015 --> 00:06:47,295
And that can produce
151
00:06:47,551 --> 00:06:50,367
Long delays when your vram is exhausted
152
00:06:50,879 --> 00:06:54,975
But if you have a very large image like this like the 2000
153
00:06:56,255 --> 00:06:58,303
You can actually get very slow case sample
154
00:06:59,071 --> 00:07:00,863
Iterations the step
155
00:07:01,119 --> 00:07:02,143
What's the refrigerator iterations
156
00:07:03,935 --> 00:07:05,983
This iterations can run very very slowly
157
00:07:06,751 --> 00:07:10,079
If your vram is exhausted which is why what happened here
158
00:07:10,335 --> 00:07:12,895
Now if you have a very powerful graphics card
159
00:07:13,151 --> 00:07:14,943
What I like you to do is to
160
00:07:16,223 --> 00:07:17,759
I'll include this particular
161
00:07:18,015 --> 00:07:22,111
Ask him and maybe what you might want to do is just to use it for inspiration
162
00:07:22,623 --> 00:07:27,743
We're not supposed to go beyond one megapixel really we're not supposed to
163
00:07:28,255 --> 00:07:30,047
In this particular design
164
00:07:30,303 --> 00:07:32,095
Goes beyond one megapixel
165
00:07:32,607 --> 00:07:36,703
Will it rain is supposed to use the refiner model with the same sampler
166
00:07:37,215 --> 00:07:39,263
As as the base model
167
00:07:39,519 --> 00:07:41,311
And this particular
168
00:07:43,615 --> 00:07:45,151
So it's it's a rule break
169
00:07:45,407 --> 00:07:46,687
This game is one
170
00:07:46,943 --> 00:07:47,967
I just want to
171
00:07:48,223 --> 00:07:49,503
Shave this with you
172
00:07:49,759 --> 00:07:51,295
Play get you to think in terms of
173
00:07:52,063 --> 00:07:53,599
You know where can I break the rules
174
00:07:53,855 --> 00:07:56,671
We know how it's supposed to work we're supposed to do
175
00:07:56,927 --> 00:07:59,743
Do most of the work with the with the base model
176
00:07:59,999 --> 00:08:00,767
And then
177
00:08:01,023 --> 00:08:03,327
With a base model here and then we work
178
00:08:03,583 --> 00:08:04,607
Quit the refin
179
00:08:05,119 --> 00:08:06,143
I also like
180
00:08:06,399 --> 00:08:08,191
The fact with
181
00:08:11,263 --> 00:08:13,055
Refiner the bass here
182
00:08:16,127 --> 00:08:18,943
As easy as it is to follow it is breaking the rules
183
00:08:19,199 --> 00:08:21,759
We're using the same sampler for the bass
184
00:08:22,527 --> 00:08:25,087
For the output
185
00:08:26,111 --> 00:08:30,207
The design is something similar to automatically 11:11
186
00:08:31,231 --> 00:08:32,767
Maybe not the best
187
00:08:33,023 --> 00:08:33,791
What's idea
188
00:08:34,303 --> 00:08:36,095
But maybe a good idea I don't know
189
00:08:36,351 --> 00:08:38,655
Is a very low D noise level here which
190
00:08:39,167 --> 00:08:43,007
Is allowing us to retain much of the detail from the original image
191
00:08:44,543 --> 00:08:46,079
Not in the in the final
192
00:08:46,335 --> 00:08:47,359
Image with the ref
193
00:08:48,383 --> 00:08:52,479
I actually like this and when I came across this image when I was planning this lecture
194
00:08:53,247 --> 00:08:55,295
I thought to myself you don't have forgotten about this
195
00:08:55,551 --> 00:08:58,623
I forgot about this somewhat unusual scheme
196
00:08:58,879 --> 00:09:00,671
And the way that it just seemed to work
197
00:09:01,695 --> 00:09:02,975
In many ways
198
00:09:03,231 --> 00:09:04,767
What's a lot better than the
199
00:09:06,047 --> 00:09:07,071
Which was supposed to use
200
00:09:07,583 --> 00:09:09,119
So I don't want you to have this
201
00:09:09,887 --> 00:09:10,655
Play around with it
202
00:09:11,167 --> 00:09:13,983
See how it differs from the schemes were supposed to use
203
00:09:14,495 --> 00:09:15,775
And maybe
204
00:09:16,543 --> 00:09:21,151
Maybe playoff it a little bit I like the idea of getting ideas from other users
205
00:09:21,407 --> 00:09:23,711
And sometimes being a little bit surprised by the results
206
00:09:24,479 --> 00:09:30,111
So this one is going to be available for you I think I will have a similar prompt or the same prompt
207
00:09:32,671 --> 00:09:33,695
We'll see what we can get
14238
Can't find what you're looking for?
Get subtitles in any language from opensubtitles.com, and translate them here.
Afrikaans
Akan
Albanian
Amharic
Arabic
Armenian
Azerbaijani
Basque
Belarusian
Bemba
Bengali
Bihari
Bosnian
Breton
Bulgarian
Cambodian
Catalan
Cebuano
Cherokee
Chichewa
Chinese (Simplified)
Chinese (Traditional)
Corsican
Croatian
Czech
Danish
Dutch
English
Esperanto
Estonian
Faroese
Finnish
Galician
Georgian
German
Greek
Guarani
Haitian Creole
Hausa
Hawaiian
Hebrew
Hmong
Hungarian
Icelandic
Indonesian
Irish
Italian
Japanese
Kazakh
Kinyarwanda
Kirundi
Kongo
Korean
Krio (Sierra Leone)
Kurdish
Kyrgyz
Laothian
Latin
Latvian
Lithuanian
Luganda
Luo
Luxembourgish
Macedonian
Malagasy
Malay
Maltese
Maori
Mauritian Creole
Moldavian
Mongolian
Myanmar (Burmese)
Montenegrin
Nepali
Norwegian
Pashto
Persian
Polish
Portuguese (Brazil)
Portuguese (Portugal)
Quechua
Romanian
Romansh
Russian
Samoan
Scots Gaelic
Serbian
Sesotho
Setswana
Seychellois Creole
Shona
Sindhi
Sinhalese
Slovak
Slovenian
Somali
Spanish
Swedish
Tajik
Thai
Tigrinya
Tonga
Turkish
Turkmen
Ukrainian
Uzbek
Welsh
Wolof