Not a member of Pastebin yet?
Sign Up,
it unlocks many cool features!
- 1
- 00:00:00,000 --> 00:00:02,030
- remember a few months ago where I
- 2
- 00:00:02,030 --> 00:00:02,040
- remember a few months ago where I
- 3
- 00:00:02,040 --> 00:00:03,649
- remember a few months ago where I
- showcased some really impressive air
- 4
- 00:00:03,649 --> 00:00:03,659
- showcased some really impressive air
- 5
- 00:00:03,659 --> 00:00:05,749
- showcased some really impressive air
- using dreambooth that can copy any
- 6
- 00:00:05,749 --> 00:00:05,759
- using dreambooth that can copy any
- 7
- 00:00:05,759 --> 00:00:08,089
- using dreambooth that can copy any
- subject or any art style and by using
- 8
- 00:00:08,089 --> 00:00:08,099
- subject or any art style and by using
- 9
- 00:00:08,099 --> 00:00:10,669
- subject or any art style and by using
- some really good fine-tuned models you
- 10
- 00:00:10,669 --> 00:00:10,679
- some really good fine-tuned models you
- 11
- 00:00:10,679 --> 00:00:12,650
- some really good fine-tuned models you
- can generate some incredible art which
- 12
- 00:00:12,650 --> 00:00:12,660
- can generate some incredible art which
- 13
- 00:00:12,660 --> 00:00:15,470
- can generate some incredible art which
- is text well four months later people
- 14
- 00:00:15,470 --> 00:00:15,480
- is text well four months later people
- 15
- 00:00:15,480 --> 00:00:17,570
- is text well four months later people
- somehow made everything look so much
- 16
- 00:00:17,570 --> 00:00:17,580
- somehow made everything look so much
- 17
- 00:00:17,580 --> 00:00:19,730
- somehow made everything look so much
- better from the extreme realism where
- 18
- 00:00:19,730 --> 00:00:19,740
- better from the extreme realism where
- 19
- 00:00:19,740 --> 00:00:21,410
- better from the extreme realism where
- the images just look like real people
- 20
- 00:00:21,410 --> 00:00:21,420
- the images just look like real people
- 21
- 00:00:21,420 --> 00:00:23,150
- the images just look like real people
- without the fake image glow that other
- 22
- 00:00:23,150 --> 00:00:23,160
- without the fake image glow that other
- 23
- 00:00:23,160 --> 00:00:25,670
- without the fake image glow that other
- AI will create or illustrations that are
- 24
- 00:00:25,670 --> 00:00:25,680
- AI will create or illustrations that are
- 25
- 00:00:25,680 --> 00:00:27,650
- AI will create or illustrations that are
- so much more coherent and stylistic
- 26
- 00:00:27,650 --> 00:00:27,660
- so much more coherent and stylistic
- 27
- 00:00:27,660 --> 00:00:29,330
- so much more coherent and stylistic
- making most of the ways that people
- 28
- 00:00:29,330 --> 00:00:29,340
- making most of the ways that people
- 29
- 00:00:29,340 --> 00:00:31,730
- making most of the ways that people
- suggested on how to identify AI are
- 30
- 00:00:31,730 --> 00:00:31,740
- suggested on how to identify AI are
- 31
- 00:00:31,740 --> 00:00:33,709
- suggested on how to identify AI are
- unusable especially with these new
- 32
- 00:00:33,709 --> 00:00:33,719
- unusable especially with these new
- 33
- 00:00:33,719 --> 00:00:35,209
- unusable especially with these new
- techniques I just want to put a
- 34
- 00:00:35,209 --> 00:00:35,219
- techniques I just want to put a
- 35
- 00:00:35,219 --> 00:00:37,010
- techniques I just want to put a
- disclaimer on this video that this does
- 36
- 00:00:37,010 --> 00:00:37,020
- disclaimer on this video that this does
- 37
- 00:00:37,020 --> 00:00:38,990
- disclaimer on this video that this does
- go down the in Kenny Valley of mine
- 38
- 00:00:38,990 --> 00:00:39,000
- go down the in Kenny Valley of mine
- 39
- 00:00:39,000 --> 00:00:41,030
- go down the in Kenny Valley of mine
- where there is gonna be even more art
- 40
- 00:00:41,030 --> 00:00:41,040
- where there is gonna be even more art
- 41
- 00:00:41,040 --> 00:00:42,889
- where there is gonna be even more art
- style copying and potential for deep
- 42
- 00:00:42,889 --> 00:00:42,899
- style copying and potential for deep
- 43
- 00:00:42,899 --> 00:00:44,690
- style copying and potential for deep
- fake images while everything is still
- 44
- 00:00:44,690 --> 00:00:44,700
- fake images while everything is still
- 45
- 00:00:44,700 --> 00:00:46,850
- fake images while everything is still
- improving over time I am still figuring
- 46
- 00:00:46,850 --> 00:00:46,860
- improving over time I am still figuring
- 47
- 00:00:46,860 --> 00:00:48,830
- improving over time I am still figuring
- out how I feel about all these but I
- 48
- 00:00:48,830 --> 00:00:48,840
- out how I feel about all these but I
- 49
- 00:00:48,840 --> 00:00:50,450
- out how I feel about all these but I
- also think everyone else deserves to
- 50
- 00:00:50,450 --> 00:00:50,460
- also think everyone else deserves to
- 51
- 00:00:50,460 --> 00:00:52,190
- also think everyone else deserves to
- learn about this because this is a
- 52
- 00:00:52,190 --> 00:00:52,200
- learn about this because this is a
- 53
- 00:00:52,200 --> 00:00:54,410
- learn about this because this is a
- technological Advance we inevitably have
- 54
- 00:00:54,410 --> 00:00:54,420
- technological Advance we inevitably have
- 55
- 00:00:54,420 --> 00:00:57,170
- technological Advance we inevitably have
- to face to quote tweet AI will not
- 56
- 00:00:57,170 --> 00:00:57,180
- to face to quote tweet AI will not
- 57
- 00:00:57,180 --> 00:00:59,750
- to face to quote tweet AI will not
- replace you a person using AIO and this
- 58
- 00:00:59,750 --> 00:00:59,760
- replace you a person using AIO and this
- 59
- 00:00:59,760 --> 00:01:01,850
- replace you a person using AIO and this
- is soon going to be a world that we live
- 60
- 00:01:01,850 --> 00:01:01,860
- is soon going to be a world that we live
- 61
- 00:01:01,860 --> 00:01:03,290
- is soon going to be a world that we live
- in so maybe getting a better
- 62
- 00:01:03,290 --> 00:01:03,300
- in so maybe getting a better
- 63
- 00:01:03,300 --> 00:01:05,270
- in so maybe getting a better
- understanding of how everything works so
- 64
- 00:01:05,270 --> 00:01:05,280
- understanding of how everything works so
- 65
- 00:01:05,280 --> 00:01:07,010
- understanding of how everything works so
- you don't get left behind is pretty
- 66
- 00:01:07,010 --> 00:01:07,020
- you don't get left behind is pretty
- 67
- 00:01:07,020 --> 00:01:09,230
- you don't get left behind is pretty
- important too anyways to get to this
- 68
- 00:01:09,230 --> 00:01:09,240
- important too anyways to get to this
- 69
- 00:01:09,240 --> 00:01:10,730
- important too anyways to get to this
- level of generated realism or
- 70
- 00:01:10,730 --> 00:01:10,740
- level of generated realism or
- 71
- 00:01:10,740 --> 00:01:12,830
- level of generated realism or
- illustration as Rome wasn't built in the
- 72
- 00:01:12,830 --> 00:01:12,840
- illustration as Rome wasn't built in the
- 73
- 00:01:12,840 --> 00:01:14,570
- illustration as Rome wasn't built in the
- day we have to go back and start from
- 74
- 00:01:14,570 --> 00:01:14,580
- day we have to go back and start from
- 75
- 00:01:14,580 --> 00:01:16,789
- day we have to go back and start from
- the ideas of dream booth and find two
- 76
- 00:01:16,789 --> 00:01:16,799
- the ideas of dream booth and find two
- 77
- 00:01:16,799 --> 00:01:18,410
- the ideas of dream booth and find two
- models which I mentioned in a previous
- 78
- 00:01:18,410 --> 00:01:18,420
- models which I mentioned in a previous
- 79
- 00:01:18,420 --> 00:01:20,570
- models which I mentioned in a previous
- video check it out if you haven't since
- 80
- 00:01:20,570 --> 00:01:20,580
- video check it out if you haven't since
- 81
- 00:01:20,580 --> 00:01:22,010
- video check it out if you haven't since
- stream Booth is good at learning
- 82
- 00:01:22,010 --> 00:01:22,020
- stream Booth is good at learning
- 83
- 00:01:22,020 --> 00:01:23,510
- stream Booth is good at learning
- literally anything with a very small
- 84
- 00:01:23,510 --> 00:01:23,520
- literally anything with a very small
- 85
- 00:01:23,520 --> 00:01:25,249
- literally anything with a very small
- amount of data set people started using
- 86
- 00:01:25,249 --> 00:01:25,259
- amount of data set people started using
- 87
- 00:01:25,259 --> 00:01:27,109
- amount of data set people started using
- dreamboots to fine tune on an art style
- 88
- 00:01:27,109 --> 00:01:27,119
- dreamboots to fine tune on an art style
- 89
- 00:01:27,119 --> 00:01:29,330
- dreamboots to fine tune on an art style
- or an object which makes fine-tuning on
- 90
- 00:01:29,330 --> 00:01:29,340
- or an object which makes fine-tuning on
- 91
- 00:01:29,340 --> 00:01:31,130
- or an object which makes fine-tuning on
- the original saber diffusion model is
- 92
- 00:01:31,130 --> 00:01:31,140
- the original saber diffusion model is
- 93
- 00:01:31,140 --> 00:01:32,749
- the original saber diffusion model is
- kind of obsolete because it takes so
- 94
- 00:01:32,749 --> 00:01:32,759
- kind of obsolete because it takes so
- 95
- 00:01:32,759 --> 00:01:34,730
- kind of obsolete because it takes so
- much more images training time and
- 96
- 00:01:34,730 --> 00:01:34,740
- much more images training time and
- 97
- 00:01:34,740 --> 00:01:37,010
- much more images training time and
- computational costs to do something that
- 98
- 00:01:37,010 --> 00:01:37,020
- computational costs to do something that
- 99
- 00:01:37,020 --> 00:01:39,170
- computational costs to do something that
- is only slightly better so some people
- 100
- 00:01:39,170 --> 00:01:39,180
- is only slightly better so some people
- 101
- 00:01:39,180 --> 00:01:41,149
- is only slightly better so some people
- in the community got really creative and
- 102
- 00:01:41,149 --> 00:01:41,159
- in the community got really creative and
- 103
- 00:01:41,159 --> 00:01:42,710
- in the community got really creative and
- started merging dreamboost and other
- 104
- 00:01:42,710 --> 00:01:42,720
- started merging dreamboost and other
- 105
- 00:01:42,720 --> 00:01:44,390
- started merging dreamboost and other
- fine-tuned models together and
- 106
- 00:01:44,390 --> 00:01:44,400
- fine-tuned models together and
- 107
- 00:01:44,400 --> 00:01:46,190
- fine-tuned models together and
- discovered that it works surprisingly
- 108
- 00:01:46,190 --> 00:01:46,200
- discovered that it works surprisingly
- 109
- 00:01:46,200 --> 00:01:47,990
- discovered that it works surprisingly
- well that pretty much started a whole
- 110
- 00:01:47,990 --> 00:01:48,000
- well that pretty much started a whole
- 111
- 00:01:48,000 --> 00:01:50,090
- well that pretty much started a whole
- new wave of merging models together to
- 112
- 00:01:50,090 --> 00:01:50,100
- new wave of merging models together to
- 113
- 00:01:50,100 --> 00:01:51,469
- new wave of merging models together to
- create something that will contain the
- 114
- 00:01:51,469 --> 00:01:51,479
- create something that will contain the
- 115
- 00:01:51,479 --> 00:01:53,569
- create something that will contain the
- best parts of the merge models however
- 116
- 00:01:53,569 --> 00:01:53,579
- best parts of the merge models however
- 117
- 00:01:53,579 --> 00:01:56,510
- best parts of the merge models however
- merging models was not a New Concept it
- 118
- 00:01:56,510 --> 00:01:56,520
- merging models was not a New Concept it
- 119
- 00:01:56,520 --> 00:01:58,550
- merging models was not a New Concept it
- existed pretty early on but because it
- 120
- 00:01:58,550 --> 00:01:58,560
- existed pretty early on but because it
- 121
- 00:01:58,560 --> 00:02:00,410
- existed pretty early on but because it
- is much more complicated than training
- 122
- 00:02:00,410 --> 00:02:00,420
- is much more complicated than training
- 123
- 00:02:00,420 --> 00:02:01,789
- is much more complicated than training
- or fine-tuning since you are just
- 124
- 00:02:01,789 --> 00:02:01,799
- or fine-tuning since you are just
- 125
- 00:02:01,799 --> 00:02:03,350
- or fine-tuning since you are just
- directly messing with the weights within
- 126
- 00:02:03,350 --> 00:02:03,360
- directly messing with the weights within
- 127
- 00:02:03,360 --> 00:02:04,969
- directly messing with the weights within
- the model architecture people didn't
- 128
- 00:02:04,969 --> 00:02:04,979
- the model architecture people didn't
- 129
- 00:02:04,979 --> 00:02:06,770
- the model architecture people didn't
- really use it when dreamboots just came
- 130
- 00:02:06,770 --> 00:02:06,780
- really use it when dreamboots just came
- 131
- 00:02:06,780 --> 00:02:09,050
- really use it when dreamboots just came
- out and made everything so easy But as
- 132
- 00:02:09,050 --> 00:02:09,060
- out and made everything so easy But as
- 133
- 00:02:09,060 --> 00:02:10,850
- out and made everything so easy But as
- time passed the more trial and error
- 134
- 00:02:10,850 --> 00:02:10,860
- time passed the more trial and error
- 135
- 00:02:10,860 --> 00:02:12,589
- time passed the more trial and error
- research was able to be made so that
- 136
- 00:02:12,589 --> 00:02:12,599
- research was able to be made so that
- 137
- 00:02:12,599 --> 00:02:14,570
- research was able to be made so that
- good merges began to emerge around the
- 138
- 00:02:14,570 --> 00:02:14,580
- good merges began to emerge around the
- 139
- 00:02:14,580 --> 00:02:16,309
- good merges began to emerge around the
- middle of January and became popular
- 140
- 00:02:16,309 --> 00:02:16,319
- middle of January and became popular
- 141
- 00:02:16,319 --> 00:02:18,110
- middle of January and became popular
- ever since there is a really good
- 142
- 00:02:18,110 --> 00:02:18,120
- ever since there is a really good
- 143
- 00:02:18,120 --> 00:02:19,850
- ever since there is a really good
- explanation of how different types of
- 144
- 00:02:19,850 --> 00:02:19,860
- explanation of how different types of
- 145
- 00:02:19,860 --> 00:02:21,530
- explanation of how different types of
- merging works I'll link it down in the
- 146
- 00:02:21,530 --> 00:02:21,540
- merging works I'll link it down in the
- 147
- 00:02:21,540 --> 00:02:23,150
- merging works I'll link it down in the
- description I'll just give a very simple
- 148
- 00:02:23,150 --> 00:02:23,160
- description I'll just give a very simple
- 149
- 00:02:23,160 --> 00:02:25,250
- description I'll just give a very simple
- explanation that classic merging Works
- 150
- 00:02:25,250 --> 00:02:25,260
- explanation that classic merging Works
- 151
- 00:02:25,260 --> 00:02:27,110
- explanation that classic merging Works
- basically by adding or calculating the
- 152
- 00:02:27,110 --> 00:02:27,120
- basically by adding or calculating the
- 153
- 00:02:27,120 --> 00:02:28,850
- basically by adding or calculating the
- differences between the weights so that
- 154
- 00:02:28,850 --> 00:02:28,860
- differences between the weights so that
- 155
- 00:02:28,860 --> 00:02:30,830
- differences between the weights so that
- the combined model would uniformly share
- 156
- 00:02:30,830 --> 00:02:30,840
- the combined model would uniformly share
- 157
- 00:02:30,840 --> 00:02:32,390
- the combined model would uniformly share
- the key features that each of them has
- 158
- 00:02:32,390 --> 00:02:32,400
- the key features that each of them has
- 159
- 00:02:32,400 --> 00:02:34,490
- the key features that each of them has
- while block merging is another type of
- 160
- 00:02:34,490 --> 00:02:34,500
- while block merging is another type of
- 161
- 00:02:34,500 --> 00:02:36,650
- while block merging is another type of
- merging which allows you to specify a
- 162
- 00:02:36,650 --> 00:02:36,660
- merging which allows you to specify a
- 163
- 00:02:36,660 --> 00:02:38,449
- merging which allows you to specify a
- particular layer of the model so it
- 164
- 00:02:38,449 --> 00:02:38,459
- particular layer of the model so it
- 165
- 00:02:38,459 --> 00:02:40,190
- particular layer of the model so it
- gives you a lot more find control on
- 166
- 00:02:40,190 --> 00:02:40,200
- gives you a lot more find control on
- 167
- 00:02:40,200 --> 00:02:42,470
- gives you a lot more find control on
- what features you combine lock merging
- 168
- 00:02:42,470 --> 00:02:42,480
- what features you combine lock merging
- 169
- 00:02:42,480 --> 00:02:44,750
- what features you combine lock merging
- is extremely useful as some part of the
- 170
- 00:02:44,750 --> 00:02:44,760
- is extremely useful as some part of the
- 171
- 00:02:44,760 --> 00:02:46,550
- is extremely useful as some part of the
- layers is pretty much equivalent to an
- 172
- 00:02:46,550 --> 00:02:46,560
- layers is pretty much equivalent to an
- 173
- 00:02:46,560 --> 00:02:48,650
- layers is pretty much equivalent to an
- abstract feature such as hands so
- 174
- 00:02:48,650 --> 00:02:48,660
- abstract feature such as hands so
- 175
- 00:02:48,660 --> 00:02:49,970
- abstract feature such as hands so
- merging a model that is good at
- 176
- 00:02:49,970 --> 00:02:49,980
- merging a model that is good at
- 177
- 00:02:49,980 --> 00:02:51,650
- merging a model that is good at
- generating hands with another that is
- 178
- 00:02:51,650 --> 00:02:51,660
- generating hands with another that is
- 179
- 00:02:51,660 --> 00:02:53,449
- generating hands with another that is
- not will pretty much guarantee their
- 180
- 00:02:53,449 --> 00:02:53,459
- not will pretty much guarantee their
- 181
- 00:02:53,459 --> 00:02:55,309
- not will pretty much guarantee their
- baby model will be able to generate good
- 182
- 00:02:55,309 --> 00:02:55,319
- baby model will be able to generate good
- 183
- 00:02:55,319 --> 00:02:57,770
- baby model will be able to generate good
- hand features so now since we can merge
- 184
- 00:02:57,770 --> 00:02:57,780
- hand features so now since we can merge
- 185
- 00:02:57,780 --> 00:03:00,229
- hand features so now since we can merge
- models together however we want then and
- 186
- 00:03:00,229 --> 00:03:00,239
- models together however we want then and
- 187
- 00:03:00,239 --> 00:03:01,790
- models together however we want then and
- there is no stopping from merging a
- 188
- 00:03:01,790 --> 00:03:01,800
- there is no stopping from merging a
- 189
- 00:03:01,800 --> 00:03:04,009
- there is no stopping from merging a
- merge model with another merge model to
- 190
- 00:03:04,009 --> 00:03:04,019
- merge model with another merge model to
- 191
- 00:03:04,019 --> 00:03:05,630
- merge model with another merge model to
- create a merge model that is used to
- 192
- 00:03:05,630 --> 00:03:05,640
- create a merge model that is used to
- 193
- 00:03:05,640 --> 00:03:08,030
- create a merge model that is used to
- merge with another model so a mix is
- 194
- 00:03:08,030 --> 00:03:08,040
- merge with another model so a mix is
- 195
- 00:03:08,040 --> 00:03:10,309
- merge with another model so a mix is
- born in some of the most popular mixes
- 196
- 00:03:10,309 --> 00:03:10,319
- born in some of the most popular mixes
- 197
- 00:03:10,319 --> 00:03:12,410
- born in some of the most popular mixes
- they will include mixed recipes which
- 198
- 00:03:12,410 --> 00:03:12,420
- they will include mixed recipes which
- 199
- 00:03:12,420 --> 00:03:13,790
- they will include mixed recipes which
- would tell you how they have merged a
- 200
- 00:03:13,790 --> 00:03:13,800
- would tell you how they have merged a
- 201
- 00:03:13,800 --> 00:03:15,290
- would tell you how they have merged a
- variety of different models to get
- 202
- 00:03:15,290 --> 00:03:15,300
- variety of different models to get
- 203
- 00:03:15,300 --> 00:03:17,270
- variety of different models to get
- something that can generate details you
- 204
- 00:03:17,270 --> 00:03:17,280
- something that can generate details you
- 205
- 00:03:17,280 --> 00:03:18,890
- something that can generate details you
- will not be able to achieve with fine
- 206
- 00:03:18,890 --> 00:03:18,900
- will not be able to achieve with fine
- 207
- 00:03:18,900 --> 00:03:20,930
- will not be able to achieve with fine
- tuning and there are mixes such as the
- 208
- 00:03:20,930 --> 00:03:20,940
- tuning and there are mixes such as the
- 209
- 00:03:20,940 --> 00:03:22,910
- tuning and there are mixes such as the
- orange mixes which is one of the most
- 210
- 00:03:22,910 --> 00:03:22,920
- orange mixes which is one of the most
- 211
- 00:03:22,920 --> 00:03:24,830
- orange mixes which is one of the most
- popular illustration mix collections
- 212
- 00:03:24,830 --> 00:03:24,840
- popular illustration mix collections
- 213
- 00:03:24,840 --> 00:03:26,869
- popular illustration mix collections
- which you can try out but here comes
- 214
- 00:03:26,869 --> 00:03:26,879
- which you can try out but here comes
- 215
- 00:03:26,879 --> 00:03:29,149
- which you can try out but here comes
- Laura a new approach released two months
- 216
- 00:03:29,149 --> 00:03:29,159
- Laura a new approach released two months
- 217
- 00:03:29,159 --> 00:03:30,949
- Laura a new approach released two months
- after a dream Booth it was originally
- 218
- 00:03:30,949 --> 00:03:30,959
- after a dream Booth it was originally
- 219
- 00:03:30,959 --> 00:03:32,509
- after a dream Booth it was originally
- made with the purpose to reduce the
- 220
- 00:03:32,509 --> 00:03:32,519
- made with the purpose to reduce the
- 221
- 00:03:32,519 --> 00:03:34,309
- made with the purpose to reduce the
- storage in the time of fine-tuning large
- 222
- 00:03:34,309 --> 00:03:34,319
- storage in the time of fine-tuning large
- 223
- 00:03:34,319 --> 00:03:36,530
- storage in the time of fine-tuning large
- language models but some people saw its
- 224
- 00:03:36,530 --> 00:03:36,540
- language models but some people saw its
- 225
- 00:03:36,540 --> 00:03:38,030
- language models but some people saw its
- potential and it was implemented into
- 226
- 00:03:38,030 --> 00:03:38,040
- potential and it was implemented into
- 227
- 00:03:38,040 --> 00:03:39,949
- potential and it was implemented into
- Sable diffusion to create something even
- 228
- 00:03:39,949 --> 00:03:39,959
- Sable diffusion to create something even
- 229
- 00:03:39,959 --> 00:03:42,229
- Sable diffusion to create something even
- more phenomenal than dreambooth to give
- 230
- 00:03:42,229 --> 00:03:42,239
- more phenomenal than dreambooth to give
- 231
- 00:03:42,239 --> 00:03:44,330
- more phenomenal than dreambooth to give
- you an idea of how Laura works usually
- 232
- 00:03:44,330 --> 00:03:44,340
- you an idea of how Laura works usually
- 233
- 00:03:44,340 --> 00:03:46,309
- you an idea of how Laura works usually
- any fine tuning or dream Booth would
- 234
- 00:03:46,309 --> 00:03:46,319
- any fine tuning or dream Booth would
- 235
- 00:03:46,319 --> 00:03:48,470
- any fine tuning or dream Booth would
- literally take your 7gb model and
- 236
- 00:03:48,470 --> 00:03:48,480
- literally take your 7gb model and
- 237
- 00:03:48,480 --> 00:03:51,170
- literally take your 7gb model and
- fine-tune the entire 7gb model but
- 238
- 00:03:51,170 --> 00:03:51,180
- fine-tune the entire 7gb model but
- 239
- 00:03:51,180 --> 00:03:53,270
- fine-tune the entire 7gb model but
- instead Laura insert new layers into the
- 240
- 00:03:53,270 --> 00:03:53,280
- instead Laura insert new layers into the
- 241
- 00:03:53,280 --> 00:03:55,130
- instead Laura insert new layers into the
- architecture and uses those few layers
- 242
- 00:03:55,130 --> 00:03:55,140
- architecture and uses those few layers
- 243
- 00:03:55,140 --> 00:03:56,449
- architecture and uses those few layers
- to learn the goal of your fine tune
- 244
- 00:03:56,449 --> 00:03:56,459
- to learn the goal of your fine tune
- 245
- 00:03:56,459 --> 00:03:58,190
- to learn the goal of your fine tune
- functionally it is very similar to
- 246
- 00:03:58,190 --> 00:03:58,200
- functionally it is very similar to
- 247
- 00:03:58,200 --> 00:03:59,750
- functionally it is very similar to
- textual inversion so you train a trigger
- 248
- 00:03:59,750 --> 00:03:59,760
- textual inversion so you train a trigger
- 249
- 00:03:59,760 --> 00:04:01,610
- textual inversion so you train a trigger
- forward with a subject for the AI to
- 250
- 00:04:01,610 --> 00:04:01,620
- forward with a subject for the AI to
- 251
- 00:04:01,620 --> 00:04:03,410
- forward with a subject for the AI to
- learn how it looks like then Laura can
- 252
- 00:04:03,410 --> 00:04:03,420
- learn how it looks like then Laura can
- 253
- 00:04:03,420 --> 00:04:05,330
- learn how it looks like then Laura can
- regenerate the idea or the object with
- 254
- 00:04:05,330 --> 00:04:05,340
- regenerate the idea or the object with
- 255
- 00:04:05,340 --> 00:04:07,009
- regenerate the idea or the object with
- the accuracy much better than textual
- 256
- 00:04:07,009 --> 00:04:07,019
- the accuracy much better than textual
- 257
- 00:04:07,019 --> 00:04:08,930
- the accuracy much better than textual
- inversion because it has a much deeper
- 258
- 00:04:08,930 --> 00:04:08,940
- inversion because it has a much deeper
- 259
- 00:04:08,940 --> 00:04:10,729
- inversion because it has a much deeper
- influence in the model even though its
- 260
- 00:04:10,729 --> 00:04:10,739
- influence in the model even though its
- 261
- 00:04:10,739 --> 00:04:12,410
- influence in the model even though its
- size is slightly larger than textual
- 262
- 00:04:12,410 --> 00:04:12,420
- size is slightly larger than textual
- 263
- 00:04:12,420 --> 00:04:14,210
- size is slightly larger than textual
- inversion it is still much smaller than
- 264
- 00:04:14,210 --> 00:04:14,220
- inversion it is still much smaller than
- 265
- 00:04:14,220 --> 00:04:16,550
- inversion it is still much smaller than
- a fine-tuned dreamboost model Laura has
- 266
- 00:04:16,550 --> 00:04:16,560
- a fine-tuned dreamboost model Laura has
- 267
- 00:04:16,560 --> 00:04:18,830
- a fine-tuned dreamboost model Laura has
- some very clever maths backing it up on
- 268
- 00:04:18,830 --> 00:04:18,840
- some very clever maths backing it up on
- 269
- 00:04:18,840 --> 00:04:20,629
- some very clever maths backing it up on
- why it works so if you want to learn
- 270
- 00:04:20,629 --> 00:04:20,639
- why it works so if you want to learn
- 271
- 00:04:20,639 --> 00:04:22,310
- why it works so if you want to learn
- more about it check out this koiboy
- 272
- 00:04:22,310 --> 00:04:22,320
- more about it check out this koiboy
- 273
- 00:04:22,320 --> 00:04:24,230
- more about it check out this koiboy
- video for a technical explanation and
- 274
- 00:04:24,230 --> 00:04:24,240
- video for a technical explanation and
- 275
- 00:04:24,240 --> 00:04:26,270
- video for a technical explanation and
- comparison with all the other methods
- 276
- 00:04:26,270 --> 00:04:26,280
- comparison with all the other methods
- 277
- 00:04:26,280 --> 00:04:28,670
- comparison with all the other methods
- things start to get weirder here though
- 278
- 00:04:28,670 --> 00:04:28,680
- things start to get weirder here though
- 279
- 00:04:28,680 --> 00:04:30,530
- things start to get weirder here though
- some big brain individuals somehow
- 280
- 00:04:30,530 --> 00:04:30,540
- some big brain individuals somehow
- 281
- 00:04:30,540 --> 00:04:32,450
- some big brain individuals somehow
- figured out how to merge Laura together
- 282
- 00:04:32,450 --> 00:04:32,460
- figured out how to merge Laura together
- 283
- 00:04:32,460 --> 00:04:34,610
- figured out how to merge Laura together
- with merged models so you can merge
- 284
- 00:04:34,610 --> 00:04:34,620
- with merged models so you can merge
- 285
- 00:04:34,620 --> 00:04:36,230
- with merged models so you can merge
- Laura with the model and merge it again
- 286
- 00:04:36,230 --> 00:04:36,240
- Laura with the model and merge it again
- 287
- 00:04:36,240 --> 00:04:38,570
- Laura with the model and merge it again
- with another Laura and vice versa so
- 288
- 00:04:38,570 --> 00:04:38,580
- with another Laura and vice versa so
- 289
- 00:04:38,580 --> 00:04:40,550
- with another Laura and vice versa so
- some really stylistic art began to pop
- 290
- 00:04:40,550 --> 00:04:40,560
- some really stylistic art began to pop
- 291
- 00:04:40,560 --> 00:04:42,650
- some really stylistic art began to pop
- out where it resembles nothing like your
- 292
- 00:04:42,650 --> 00:04:42,660
- out where it resembles nothing like your
- 293
- 00:04:42,660 --> 00:04:44,930
- out where it resembles nothing like your
- typical AI generated art also the
- 294
- 00:04:44,930 --> 00:04:44,940
- typical AI generated art also the
- 295
- 00:04:44,940 --> 00:04:46,850
- typical AI generated art also the
- progress made in text generated are is
- 296
- 00:04:46,850 --> 00:04:46,860
- progress made in text generated are is
- 297
- 00:04:46,860 --> 00:04:48,830
- progress made in text generated are is
- either made by the weebs or by the horny
- 298
- 00:04:48,830 --> 00:04:48,840
- either made by the weebs or by the horny
- 299
- 00:04:48,840 --> 00:04:50,570
- either made by the weebs or by the horny
- people in real life text generated
- 300
- 00:04:50,570 --> 00:04:50,580
- people in real life text generated
- 301
- 00:04:50,580 --> 00:04:52,909
- people in real life text generated
- images are what impressed and worried me
- 302
- 00:04:52,909 --> 00:04:52,919
- images are what impressed and worried me
- 303
- 00:04:52,919 --> 00:04:54,950
- images are what impressed and worried me
- the most if you remember any text to
- 304
- 00:04:54,950 --> 00:04:54,960
- the most if you remember any text to
- 305
- 00:04:54,960 --> 00:04:57,110
- the most if you remember any text to
- image results on a human they all have
- 306
- 00:04:57,110 --> 00:04:57,120
- image results on a human they all have
- 307
- 00:04:57,120 --> 00:04:59,510
- image results on a human they all have
- this weird artificial glow or textures
- 308
- 00:04:59,510 --> 00:04:59,520
- this weird artificial glow or textures
- 309
- 00:04:59,520 --> 00:05:01,969
- this weird artificial glow or textures
- which you can instantly recognize its AI
- 310
- 00:05:01,969 --> 00:05:01,979
- which you can instantly recognize its AI
- 311
- 00:05:01,979 --> 00:05:04,490
- which you can instantly recognize its AI
- generated but with these mixes on top of
- 312
- 00:05:04,490 --> 00:05:04,500
- generated but with these mixes on top of
- 313
- 00:05:04,500 --> 00:05:06,469
- generated but with these mixes on top of
- Laura people have found the perfect
- 314
- 00:05:06,469 --> 00:05:06,479
- Laura people have found the perfect
- 315
- 00:05:06,479 --> 00:05:08,390
- Laura people have found the perfect
- combo that will make the generation look
- 316
- 00:05:08,390 --> 00:05:08,400
- combo that will make the generation look
- 317
- 00:05:08,400 --> 00:05:10,550
- combo that will make the generation look
- extremely real to the point that I had
- 318
- 00:05:10,550 --> 00:05:10,560
- extremely real to the point that I had
- 319
- 00:05:10,560 --> 00:05:12,409
- extremely real to the point that I had
- to double take when I first came across
- 320
- 00:05:12,409 --> 00:05:12,419
- to double take when I first came across
- 321
- 00:05:12,419 --> 00:05:14,210
- to double take when I first came across
- them the mixes are based on some really
- 322
- 00:05:14,210 --> 00:05:14,220
- them the mixes are based on some really
- 323
- 00:05:14,220 --> 00:05:17,090
- them the mixes are based on some really
- Sinister or Unholy models so I'll not be
- 324
- 00:05:17,090 --> 00:05:17,100
- Sinister or Unholy models so I'll not be
- 325
- 00:05:17,100 --> 00:05:18,830
- Sinister or Unholy models so I'll not be
- linking anything about it since it
- 326
- 00:05:18,830 --> 00:05:18,840
- linking anything about it since it
- 327
- 00:05:18,840 --> 00:05:21,230
- linking anything about it since it
- violates YouTube TOS but the safe for
- 328
- 00:05:21,230 --> 00:05:21,240
- violates YouTube TOS but the safe for
- 329
- 00:05:21,240 --> 00:05:23,450
- violates YouTube TOS but the safe for
- work results are mind-blowing the only
- 330
- 00:05:23,450 --> 00:05:23,460
- work results are mind-blowing the only
- 331
- 00:05:23,460 --> 00:05:25,490
- work results are mind-blowing the only
- dead giveaways of it being AI generator
- 332
- 00:05:25,490 --> 00:05:25,500
- dead giveaways of it being AI generator
- 333
- 00:05:25,500 --> 00:05:27,590
- dead giveaways of it being AI generator
- are either the clothing patterns or the
- 334
- 00:05:27,590 --> 00:05:27,600
- are either the clothing patterns or the
- 335
- 00:05:27,600 --> 00:05:30,290
- are either the clothing patterns or the
- accessories however the face the anatomy
- 336
- 00:05:30,290 --> 00:05:30,300
- accessories however the face the anatomy
- 337
- 00:05:30,300 --> 00:05:32,450
- accessories however the face the anatomy
- the lighting the shading the reflection
- 338
- 00:05:32,450 --> 00:05:32,460
- the lighting the shading the reflection
- 339
- 00:05:32,460 --> 00:05:35,090
- the lighting the shading the reflection
- it is scary how real all these look
- 340
- 00:05:35,090 --> 00:05:35,100
- it is scary how real all these look
- 341
- 00:05:35,100 --> 00:05:36,469
- it is scary how real all these look
- people would probably just need to
- 342
- 00:05:36,469 --> 00:05:36,479
- people would probably just need to
- 343
- 00:05:36,479 --> 00:05:38,450
- people would probably just need to
- figure out a merge to improve these
- 344
- 00:05:38,450 --> 00:05:38,460
- figure out a merge to improve these
- 345
- 00:05:38,460 --> 00:05:40,610
- figure out a merge to improve these
- small details like the buttons hands or
- 346
- 00:05:40,610 --> 00:05:40,620
- small details like the buttons hands or
- 347
- 00:05:40,620 --> 00:05:42,650
- small details like the buttons hands or
- even the text and it could fool people
- 348
- 00:05:42,650 --> 00:05:42,660
- even the text and it could fool people
- 349
- 00:05:42,660 --> 00:05:44,749
- even the text and it could fool people
- easily maybe the next job that's going
- 350
- 00:05:44,749 --> 00:05:44,759
- easily maybe the next job that's going
- 351
- 00:05:44,759 --> 00:05:46,070
- easily maybe the next job that's going
- to be replaced are probably the
- 352
- 00:05:46,070 --> 00:05:46,080
- to be replaced are probably the
- 353
- 00:05:46,080 --> 00:05:48,050
- to be replaced are probably the
- cosplayers and the Instagram stars or
- 354
- 00:05:48,050 --> 00:05:48,060
- cosplayers and the Instagram stars or
- 355
- 00:05:48,060 --> 00:05:49,730
- cosplayers and the Instagram stars or
- the internet will be saturated with fake
- 356
- 00:05:49,730 --> 00:05:49,740
- the internet will be saturated with fake
- 357
- 00:05:49,740 --> 00:05:51,830
- the internet will be saturated with fake
- AI generator celebrities that appeal to
- 358
- 00:05:51,830 --> 00:05:51,840
- AI generator celebrities that appeal to
- 359
- 00:05:51,840 --> 00:05:53,570
- AI generator celebrities that appeal to
- your personal likings and if you
- 360
- 00:05:53,570 --> 00:05:53,580
- your personal likings and if you
- 361
- 00:05:53,580 --> 00:05:55,129
- your personal likings and if you
- incorporate control net which I talked
- 362
- 00:05:55,129 --> 00:05:55,139
- incorporate control net which I talked
- 363
- 00:05:55,139 --> 00:05:56,990
- incorporate control net which I talked
- about in my previous video you will gain
- 364
- 00:05:56,990 --> 00:05:57,000
- about in my previous video you will gain
- 365
- 00:05:57,000 --> 00:05:58,730
- about in my previous video you will gain
- so much more control while generating
- 366
- 00:05:58,730 --> 00:05:58,740
- so much more control while generating
- 367
- 00:05:58,740 --> 00:06:01,129
- so much more control while generating
- with these AIS the Precision that you'll
- 368
- 00:06:01,129 --> 00:06:01,139
- with these AIS the Precision that you'll
- 369
- 00:06:01,139 --> 00:06:02,629
- with these AIS the Precision that you'll
- be able to generate is going to change
- 370
- 00:06:02,629 --> 00:06:02,639
- be able to generate is going to change
- 371
- 00:06:02,639 --> 00:06:05,029
- be able to generate is going to change
- the digital media landscape it is also
- 372
- 00:06:05,029 --> 00:06:05,039
- the digital media landscape it is also
- 373
- 00:06:05,039 --> 00:06:06,830
- the digital media landscape it is also
- completely possible to fine-tune these
- 374
- 00:06:06,830 --> 00:06:06,840
- completely possible to fine-tune these
- 375
- 00:06:06,840 --> 00:06:09,110
- completely possible to fine-tune these
- extreme realism models with Laura to a
- 376
- 00:06:09,110 --> 00:06:09,120
- extreme realism models with Laura to a
- 377
- 00:06:09,120 --> 00:06:10,969
- extreme realism models with Laura to a
- specific person which in this case can
- 378
- 00:06:10,969 --> 00:06:10,979
- specific person which in this case can
- 379
- 00:06:10,979 --> 00:06:13,249
- specific person which in this case can
- be potentially as dangerous as deep fake
- 380
- 00:06:13,249 --> 00:06:13,259
- be potentially as dangerous as deep fake
- 381
- 00:06:13,259 --> 00:06:14,870
- be potentially as dangerous as deep fake
- while being much more effortless to
- 382
- 00:06:14,870 --> 00:06:14,880
- while being much more effortless to
- 383
- 00:06:14,880 --> 00:06:16,969
- while being much more effortless to
- generate once a model of them exists
- 384
- 00:06:16,969 --> 00:06:16,979
- generate once a model of them exists
- 385
- 00:06:16,979 --> 00:06:18,950
- generate once a model of them exists
- comparing the amount of hours you need
- 386
- 00:06:18,950 --> 00:06:18,960
- comparing the amount of hours you need
- 387
- 00:06:18,960 --> 00:06:21,230
- comparing the amount of hours you need
- to defeat a brand new video versus 6
- 388
- 00:06:21,230 --> 00:06:21,240
- to defeat a brand new video versus 6
- 389
- 00:06:21,240 --> 00:06:22,850
- to defeat a brand new video versus 6
- minutes training of a Lora with your
- 390
- 00:06:22,850 --> 00:06:22,860
- minutes training of a Lora with your
- 391
- 00:06:22,860 --> 00:06:24,529
- minutes training of a Lora with your
- face and anyone that obtains the model
- 392
- 00:06:24,529 --> 00:06:24,539
- face and anyone that obtains the model
- 393
- 00:06:24,539 --> 00:06:26,270
- face and anyone that obtains the model
- will be able to generate anything with
- 394
- 00:06:26,270 --> 00:06:26,280
- will be able to generate anything with
- 395
- 00:06:26,280 --> 00:06:28,610
- will be able to generate anything with
- your face on it infinitely the future is
- 396
- 00:06:28,610 --> 00:06:28,620
- your face on it infinitely the future is
- 397
- 00:06:28,620 --> 00:06:30,110
- your face on it infinitely the future is
- going to be extremely worrying and
- 398
- 00:06:30,110 --> 00:06:30,120
- going to be extremely worrying and
- 399
- 00:06:30,120 --> 00:06:31,850
- going to be extremely worrying and
- terrifying for some thank you so much
- 400
- 00:06:31,850 --> 00:06:31,860
- terrifying for some thank you so much
- 401
- 00:06:31,860 --> 00:06:33,610
- terrifying for some thank you so much
- for watching a big shout out to Andrew
- 402
- 00:06:33,610 --> 00:06:33,620
- for watching a big shout out to Andrew
- 403
- 00:06:33,620 --> 00:06:38,150
- for watching a big shout out to Andrew
- lascellias Chris LeDoux Alex Marie's and
- 404
- 00:06:38,150 --> 00:06:38,160
- lascellias Chris LeDoux Alex Marie's and
- 405
- 00:06:38,160 --> 00:06:39,469
- lascellias Chris LeDoux Alex Marie's and
- many others that support me through
- 406
- 00:06:39,469 --> 00:06:39,479
- many others that support me through
- 407
- 00:06:39,479 --> 00:06:41,570
- many others that support me through
- patreon or YouTube follow my Twitter if
- 408
- 00:06:41,570 --> 00:06:41,580
- patreon or YouTube follow my Twitter if
- 409
- 00:06:41,580 --> 00:06:43,249
- patreon or YouTube follow my Twitter if
- you haven't and I'll see you all in the
- 410
- 00:06:43,249 --> 00:06:43,259
- you haven't and I'll see you all in the
- 411
- 00:06:43,259 --> 00:06:45,500
- you haven't and I'll see you all in the
- next one
Add Comment
Please, Sign In to add comment