1
00:00:00,180 --> 00:00:06,720
Hi and welcome back in this lesson, we'll take a look at generating our very own anime characters using

2
00:00:06,720 --> 00:00:07,380
style again.

3
00:00:07,830 --> 00:00:11,490
So go ahead and open up less than 24 and we'll begin to listen.

4
00:00:12,180 --> 00:00:18,000
So firstly, this lesson requires us to downgrade to TensorFlow one point one point one five.

5
00:00:18,510 --> 00:00:19,970
So let's do that.

6
00:00:20,010 --> 00:00:22,530
And to do that in cloud, it's actually quite easy.

7
00:00:22,530 --> 00:00:25,050
Just use this magic command here.

8
00:00:25,050 --> 00:00:31,470
TensorFlow Vision set it to one point x and it will automatically select and you can see it loaded that

9
00:00:31,470 --> 00:00:32,430
version of TensorFlow.

10
00:00:32,550 --> 00:00:33,300
It's quite cool.

11
00:00:33,750 --> 00:00:39,690
Otherwise, if you didn't have this line, every load TensorFlow 2.5 or 2.6, whatever little solution

12
00:00:39,690 --> 00:00:42,210
that's installed in the cooler back end.

13
00:00:42,870 --> 00:00:49,140
So now we have to clone this model, so we're going to use a pre-trained gun that's available online.

14
00:00:49,590 --> 00:00:52,000
So let's clone that from this GitHub repo.

15
00:00:52,560 --> 00:00:53,880
And that's done.

16
00:00:54,720 --> 00:00:58,160
Now we're going to download the pre-trained network, which as well.

17
00:00:58,170 --> 00:00:59,490
So let's do that.

18
00:01:00,360 --> 00:01:05,430
And then we create the function here called generate XSS from seeds.

19
00:01:05,670 --> 00:01:07,800
So let's run that as well.

20
00:01:09,090 --> 00:01:15,900
And then we're going to use this DNN library here from the tier live within us to live this well.

21
00:01:16,530 --> 00:01:21,030
And so we're going to use this function here to generate our images.

22
00:01:25,490 --> 00:01:30,470
So let's wait for these things to install an import that's still downloading the network here.

23
00:01:34,790 --> 00:01:37,760
There we go, so that's finished Typekit roughly 46 seconds.

24
00:01:38,570 --> 00:01:42,650
So now we can generate some random see, it's because remember when you're building guns, we have to

25
00:01:43,280 --> 00:01:45,050
feed it some seed information here.

26
00:01:45,560 --> 00:01:47,540
So this effect is actually quite big.

27
00:01:47,540 --> 00:01:49,430
It's sitting at six million or 10 million.

28
00:01:50,000 --> 00:01:54,530
So we generate from seeds to print it out and then we feed it to.

29
00:01:54,530 --> 00:02:02,060
It would generate the gas from seeds here, and we just use just to generate images and we get our images

30
00:02:02,060 --> 00:02:02,450
here.

31
00:02:03,050 --> 00:02:05,870
Now we can visualize other images by just running that.

32
00:02:07,790 --> 00:02:12,170
And you can see this is an anime character which we just generated.

33
00:02:12,830 --> 00:02:13,820
How cool is that?

34
00:02:14,060 --> 00:02:15,410
Let's take a look at doing it again.

35
00:02:15,830 --> 00:02:20,540
Remember, this is using a pre-trained network and this is raising at a random see that we created here.

36
00:02:21,110 --> 00:02:24,920
So if we run this again, like I just did, you'll get a new image here.

37
00:02:25,850 --> 00:02:27,320
This is pretty amazing, isn't it?

38
00:02:28,100 --> 00:02:33,830
So now let's create an image grid so we can take a look at it quite a few different, randomly generated

39
00:02:33,830 --> 00:02:34,880
anime characters.

40
00:02:36,680 --> 00:02:40,220
So all of these are randomly generated anime characters.

41
00:02:41,040 --> 00:02:49,520
We can create an intercollegiate function and some more random scenes here and then generate a few more

42
00:02:49,940 --> 00:02:50,570
images.

43
00:02:53,750 --> 00:02:58,100
However, what we're doing here, what's different is that because of this into police function that

44
00:02:58,100 --> 00:03:04,550
we've created here, we used up nobody random see just to create slight variations of the same character

45
00:03:04,880 --> 00:03:07,970
so we can move it from one system to another.

46
00:03:08,360 --> 00:03:10,220
So you can see the transitions here.

47
00:03:10,220 --> 00:03:15,380
So let's take a look at that and you can see this is quite poorly concealed moves slowly, slowly from

48
00:03:15,380 --> 00:03:17,330
this into this new character here.

49
00:03:17,930 --> 00:03:20,510
So that's it for this gun lesson.

50
00:03:21,050 --> 00:03:28,190
In future, I will probably add a short video on how this gun was created, because even though you

51
00:03:28,190 --> 00:03:33,260
do have background knowledge and how to create guns, you may not have knowledge about style gun.

52
00:03:33,260 --> 00:03:37,460
So it's probably worth putting up a video to explain this style again to you guys.

53
00:03:37,460 --> 00:03:43,490
So I'll make a note of that and added on to the course shortly, hopefully in January or February of

54
00:03:43,490 --> 00:03:44,930
2022.

55
00:03:45,680 --> 00:03:52,800
OK, so that's it for this lesson will now take a look at using cycle gun to convert horses into zebras.

56
00:03:52,820 --> 00:03:54,200
So stay tuned for that lesson.

57
00:03:54,650 --> 00:03:55,040
Thank you.