Would you like to inspect the original subtitles? These are the user uploaded subtitles that are being translated:
1
00:00:00,080 --> 00:00:03,890
We have learned that stable diffusion is a diffusion model.
2
00:00:03,890 --> 00:00:11,240
In this video we will take a closer look and we need to see what Laura's are and what checkpoints are.
3
00:00:11,270 --> 00:00:14,390
These two things are really, really important.
4
00:00:14,390 --> 00:00:19,730
So we make this like really really simple in paint this right here.
5
00:00:19,730 --> 00:00:22,460
Just think of this as stable diffusion.
6
00:00:22,460 --> 00:00:24,860
So this is the big big computer.
7
00:00:24,860 --> 00:00:27,560
This is just a program stable diffusion.
8
00:00:27,560 --> 00:00:32,360
This is trained of course on a lot of pictures.
9
00:00:32,360 --> 00:00:39,710
But this enormous diffusion model has of course more versions, more checkpoints.
10
00:00:39,740 --> 00:00:44,840
And there are different checkpoints for different stuff that we want to make.
11
00:00:44,840 --> 00:00:48,710
There is, for example, stable diffusion, Excel.
12
00:00:48,710 --> 00:00:53,480
That's basically just a model that makes really, really good pictures.
13
00:00:53,480 --> 00:00:58,040
Maybe from faces may be really, really good realistic ones.
14
00:00:58,040 --> 00:01:00,500
We will use such a checkpoint.
15
00:01:00,500 --> 00:01:05,210
So we have also stable diffusion 1.5.
16
00:01:05,210 --> 00:01:10,070
And we have of course stable diffusion 2.0 and so on.
17
00:01:10,070 --> 00:01:11,720
Just think about this.
18
00:01:11,720 --> 00:01:17,570
There are a lot of different stable diffusion models stable diffusion checkpoints.
19
00:01:17,660 --> 00:01:21,140
And they are always trained on more data.
20
00:01:21,140 --> 00:01:23,390
So stable diffusion is the big thing.
21
00:01:23,390 --> 00:01:26,660
And then we have this checkpoints stable diffusion.
22
00:01:26,660 --> 00:01:32,090
Excel is for example something that makes really really good realistic pictures.
23
00:01:32,090 --> 00:01:34,850
And it depends what you want to make.
24
00:01:34,850 --> 00:01:39,770
We have also stuff that makes really, really good anime pictures and so on.
25
00:01:39,770 --> 00:01:48,590
The important stuff is that the training of this pictures takes a lot of time and is not that specific.
26
00:01:48,740 --> 00:01:50,720
Here is a big article.
27
00:01:50,720 --> 00:01:53,000
We just take a quick look at this.
28
00:01:53,000 --> 00:01:59,180
They want to explain to you that there are a lot of different diffusion models on stable diffusion.
29
00:01:59,180 --> 00:02:08,180
So different checkpoints, stable diffusion 1.41.5 realistic version, dream shaper stable diffusion,
30
00:02:08,180 --> 00:02:09,350
Excel anything.
31
00:02:09,350 --> 00:02:10,910
Version three deliberate.
32
00:02:10,910 --> 00:02:11,870
Version two.
33
00:02:11,900 --> 00:02:15,290
This deliberate, for example, is perfect for Animas.
34
00:02:15,290 --> 00:02:18,950
Stable diffusion is perfect for realistic pictures.
35
00:02:18,950 --> 00:02:22,850
Also, this realistic version is good for realistic pictures.
36
00:02:22,850 --> 00:02:30,590
The concept here is just that every different checkpoint is trained on a bit of different pictures.
37
00:02:30,590 --> 00:02:32,930
So here we have a good example.
38
00:02:33,020 --> 00:02:38,630
We have the realistic version, we have the entry and we have the dream bar.
39
00:02:38,630 --> 00:02:44,180
You can see the realistic version makes of course realistic pictures.
40
00:02:44,180 --> 00:02:52,340
The anything version is a bit more like anime and the Dream Sharper is also in this direction and it's
41
00:02:52,340 --> 00:02:59,720
realistic painting style, so every different checkpoint is simply a different style.
42
00:03:00,080 --> 00:03:03,770
And that's basically a checkpoint in stable diffusion.
43
00:03:03,770 --> 00:03:11,810
The model version 1.4 is good in such pictures, and another model is better in other pictures.
44
00:03:11,810 --> 00:03:19,010
Generally speaking, we need to use realistic versions and they are simply trained on realistic pictures
45
00:03:19,010 --> 00:03:21,470
to get realistic outputs.
46
00:03:21,470 --> 00:03:24,290
I hope all of this make sense.
47
00:03:24,290 --> 00:03:28,190
Then we have this dream sharper that makes good animals and so on.
48
00:03:28,190 --> 00:03:34,280
The important stuff is that all of these models so even the dream sharper and so on.
49
00:03:34,280 --> 00:03:38,330
These are all ways, really, really big models.
50
00:03:38,330 --> 00:03:43,880
And we only have a handful, but we have a really, really cool solution.
51
00:03:43,880 --> 00:03:46,160
We have also Laura's.
52
00:03:46,160 --> 00:03:52,910
There is the possibility to use Laura's and to fine tune these models even further.
53
00:03:52,910 --> 00:03:54,830
Think about it this way.
54
00:03:54,830 --> 00:04:00,530
Let's say we use stable diffusion, but we want a really, really specific output.
55
00:04:00,530 --> 00:04:06,230
Maybe we want a picture that looks exactly the same in almost every picture.
56
00:04:06,710 --> 00:04:12,680
We can fine tune this stable diffusion excel with a Laura and this works really nice.
57
00:04:12,680 --> 00:04:19,130
So we work on this checkpoint and then we take a little piece of this checkpoint, this piece we call
58
00:04:19,130 --> 00:04:20,990
Laura and this Laura.
59
00:04:20,990 --> 00:04:28,220
We train on additional pictures and it's irrelevant what pictures you use.
60
00:04:28,250 --> 00:04:35,690
I trained myself, for example, a Laura on this pictures, so it's simply my face in different poses.
61
00:04:35,690 --> 00:04:43,130
And if we train a model on all of these pictures, we can create output that looks similar.
62
00:04:43,280 --> 00:04:46,130
But I didn't use stable diffusion Excel.
63
00:04:46,130 --> 00:04:50,270
I used a model that makes like animal pictures.
64
00:04:50,270 --> 00:04:54,770
This is, for example, a picture that I have made with this model.
65
00:04:54,770 --> 00:04:59,510
So first we train our model on a lot of pictures of me the.
66
00:04:59,650 --> 00:05:02,410
Model or the checkpoint was stable.
67
00:05:02,410 --> 00:05:06,910
Diffusion at Ram, sharper model and the low rise then are fine tuned.
68
00:05:06,910 --> 00:05:08,740
Lora of myself.
69
00:05:09,070 --> 00:05:11,020
That's also a picture of the model.
70
00:05:11,020 --> 00:05:15,460
And this like you can see you can make different kinds of output.
71
00:05:15,460 --> 00:05:20,890
And the cool stuff is we don't have to train our lowers ourself.
72
00:05:20,980 --> 00:05:28,000
The training of the Lora takes a bit of time, but we can search for Laura's that we like because there
73
00:05:28,000 --> 00:05:31,090
are millions of Laura's out there.
74
00:05:31,090 --> 00:05:37,930
And if we take the right Laura and we take it every time, we will get consistent characters.
75
00:05:37,930 --> 00:05:41,800
So in this video, we took a closer look at the diffusion.
76
00:05:41,800 --> 00:05:45,790
You have learned that stable diffusion is the big diffusion model.
77
00:05:45,790 --> 00:05:47,800
Then there are checkpoints.
78
00:05:47,800 --> 00:05:51,880
The checkpoints are something like the versions of stable diffusion.
79
00:05:51,880 --> 00:05:55,060
They are of course trained on different styles.
80
00:05:55,060 --> 00:05:56,770
You have a realistic one.
81
00:05:56,770 --> 00:06:01,750
So stable diffusion Excel you have the dream sharper for animals and so on.
82
00:06:01,750 --> 00:06:04,390
And they make really, really good pictures.
83
00:06:04,390 --> 00:06:06,400
But they are really, really big.
84
00:06:06,400 --> 00:06:11,680
It takes months of training for such models, for such checkpoints.
85
00:06:11,680 --> 00:06:13,540
And then we have Laura's.
86
00:06:13,570 --> 00:06:19,150
Laura's are smaller checkpoints that we can plug in in our big checkpoints.
87
00:06:19,150 --> 00:06:28,120
And with Laura's we can fine tune our models, our checkpoints, we can fine tune our Laura's ourself.
88
00:06:28,120 --> 00:06:30,070
This takes a lot of time.
89
00:06:30,070 --> 00:06:31,570
Google Colab and so on.
90
00:06:31,570 --> 00:06:34,540
And we won't do that in this course.
91
00:06:34,540 --> 00:06:40,780
But we can also take Laura's that other people have trained and we will search.
92
00:06:40,780 --> 00:06:41,410
Perfect.
93
00:06:41,410 --> 00:06:42,340
Laura's.
94
00:06:42,370 --> 00:06:45,670
That's basically the concept of stable diffusion.
95
00:06:45,670 --> 00:06:51,070
In the next video, we take a quick look at the seat and then we are ready to go.
9098
Can't find what you're looking for?
Get subtitles in any language from opensubtitles.com, and translate them here.