WEBVTT 00:00.720 --> 00:02.430 -: Welcome to "Intro to Midjourney." 00:02.430 --> 00:03.810 It's an image generation model, 00:03.810 --> 00:06.450 similar to Stable Diffusion or DALL-E, 00:06.450 --> 00:08.550 and you can see how that works here. 00:08.550 --> 00:11.850 This takes a lot of random images from the internet, 00:11.850 --> 00:14.700 like pictures of cats, adds a little bit of noise 00:14.700 --> 00:16.770 to the image until eventually it's all noise. 00:16.770 --> 00:19.530 It's just random kind of pixels on the screen. 00:19.530 --> 00:21.750 Once it trains enough on those images, 00:21.750 --> 00:24.093 and it knows what the caption and image pair is, 00:24.093 --> 00:25.770 then you can reverse the process, 00:25.770 --> 00:27.990 and you just take some random noise 00:27.990 --> 00:31.110 and then turn it into the image that matches that caption. 00:31.110 --> 00:33.960 The caption is the prompt that you're giving it. 00:33.960 --> 00:36.060 So Midjourney works a little bit different. 00:36.060 --> 00:40.410 It's actually created through an independent research lab, 00:40.410 --> 00:42.540 so it doesn't have like all the crazy funding 00:42.540 --> 00:44.619 and resources that OpenAI has. 00:44.619 --> 00:46.230 It also, you know, doesn't have all the kind 00:46.230 --> 00:48.510 of free contribution that Stable Diffusion gets 00:48.510 --> 00:50.340 from being open source, 00:50.340 --> 00:53.340 but it's still managed to grow pretty rapidly. 00:53.340 --> 00:54.660 And the reason why, I think, 00:54.660 --> 00:56.280 is through the Discord community. 00:56.280 --> 00:58.920 So in order to use the platform, 00:58.920 --> 01:00.810 you actually have to be in Discord, 01:00.810 --> 01:03.060 which is like similar to Slack if you don't know. 01:03.060 --> 01:04.860 And there are millions of people in there 01:04.860 --> 01:06.780 creating prompts every day. 01:06.780 --> 01:10.800 You know, when you pay, you can prompt the bot directly. 01:10.800 --> 01:13.410 So it's a Midjourney bot that you can DM, 01:13.410 --> 01:15.510 but also all of the prompts get posted 01:15.510 --> 01:17.580 into different channels. 01:17.580 --> 01:19.290 You can actually, if you are a paying customer, 01:19.290 --> 01:22.650 then hide your images. 01:22.650 --> 01:25.980 But it's community first by default. 01:25.980 --> 01:27.450 It's a monthly subscription, 01:27.450 --> 01:30.090 if you are a paying customer, after a free trial. 01:30.090 --> 01:32.910 So pretty simple, easy business model, 01:32.910 --> 01:34.380 and apparently they're already profitable. 01:34.380 --> 01:39.380 So, that's a very interesting, unexpected outcome. 01:39.450 --> 01:41.790 So a few things that are interesting about Midjourney. 01:41.790 --> 01:43.260 One is it has negative prompts, 01:43.260 --> 01:45.630 and, you know, that can get horrifying quickly. 01:45.630 --> 01:47.790 Like in this case, I took Peppa Pig, 01:47.790 --> 01:50.220 which is a popular cartoon my daughter likes, 01:50.220 --> 01:52.800 and then I added dash dash no. 01:52.800 --> 01:55.470 And then after a dash dash no, I put cartoon. 01:55.470 --> 01:58.860 So anything after dash dash no is a negative prompt, 01:58.860 --> 01:59.760 something you don't want. 01:59.760 --> 02:02.580 So it is giving you Peppa Pig, which is normally a cartoon, 02:02.580 --> 02:05.850 but turning it into like a photorealistic Peppa Pig. 02:05.850 --> 02:06.683 You can upscale. 02:06.683 --> 02:10.830 So, you know, after every set of image generations here, 02:10.830 --> 02:13.680 it tells you, you know, which one do you wanna upscale, 02:13.680 --> 02:16.470 or do you wanna regenerate, or do you want new versions 02:16.470 --> 02:18.150 of one of the ones that you liked? 02:18.150 --> 02:20.970 So if I clicked V4, it would generate new versions 02:20.970 --> 02:23.100 of version four, or if I click upscale, 02:23.100 --> 02:26.073 then it generates the fuller image here. 02:26.970 --> 02:28.770 Then you can also do image blending, 02:28.770 --> 02:29.850 which I think is pretty unique. 02:29.850 --> 02:32.160 You can take two images and mix 'em together. 02:32.160 --> 02:34.800 In this case, I put like a pastoral village scene 02:34.800 --> 02:38.250 from the UK and then kind of blended it 02:38.250 --> 02:41.850 with this image to make kind of a nice Peppa Pig 02:41.850 --> 02:44.550 that's not, that's in like a lovely village 02:44.550 --> 02:46.250 rather than like this horror show. 02:47.400 --> 02:50.700 The aesthetic of Midjourney is like very fantasy. 02:50.700 --> 02:52.620 So this is that pastoral village scene 02:52.620 --> 02:55.680 that I created with Midjourney as well. 02:55.680 --> 03:00.000 It's really, it looks like a video game, right? 03:00.000 --> 03:03.930 Or like a movie or, you know, kind of like one 03:03.930 --> 03:06.780 of those kind of unreal engine-generated things. 03:06.780 --> 03:10.320 And you know, this is an example of I just put in, 03:10.320 --> 03:12.720 you know, test tubes, right, and it came up with this, 03:12.720 --> 03:15.780 like it's a very fantasy kind of vision 03:15.780 --> 03:17.310 of what a test tube is. 03:17.310 --> 03:18.143 Same thing here. 03:18.143 --> 03:19.530 This is a pair of shoes 03:19.530 --> 03:23.760 that are designed in the style of "Alien Predator." 03:23.760 --> 03:26.640 So this is either a blessing or a curse 03:26.640 --> 03:29.580 depending on if you like this aesthetic. 03:29.580 --> 03:32.370 Like if you are designing AI art, 03:32.370 --> 03:34.650 then this is the type of thing 03:34.650 --> 03:37.110 that gets a lot of clicks and views, 03:37.110 --> 03:40.890 but it might not be good for every type of task. 03:40.890 --> 03:43.560 The types of tasks that I've used it for specifically, 03:43.560 --> 03:45.360 or I've seen other people use it for, 03:45.360 --> 03:46.470 one, is book art. 03:46.470 --> 03:51.000 So, you have, this is actually some illustration that I made 03:51.000 --> 03:53.580 for a book that I'm writing of like a samurai 03:53.580 --> 03:55.710 in a futuristic city. 03:55.710 --> 03:58.950 You know, this is a stock photo, right? 03:58.950 --> 04:00.570 Well, it's actually not really a stock photo. 04:00.570 --> 04:02.700 I generated it with Midjourney, 04:02.700 --> 04:04.560 but it kind of matches the type of thing 04:04.560 --> 04:07.050 that I would use on my blog for a stock photo. 04:07.050 --> 04:10.380 So the cool thing is then it's unique to me, right? 04:10.380 --> 04:13.920 As a paying customer of Midjourney, I own this image, right? 04:13.920 --> 04:15.840 Like I can use it on my blog, 04:15.840 --> 04:18.870 and I don't have to pay for like a copyright license fee. 04:18.870 --> 04:20.340 And then movie characters. 04:20.340 --> 04:22.350 So this is something I've seen a lot of people do, 04:22.350 --> 04:25.320 is they're creating like whole movies or kind of cartoons 04:25.320 --> 04:28.860 or comics using Midjourney, 04:28.860 --> 04:31.350 and it's really good at this out of the box. 04:31.350 --> 04:35.160 And you know, I think Midjourney, in general, 04:35.160 --> 04:38.460 is one of those tools that nobody expected 04:38.460 --> 04:39.630 to kind of still be in the running, 04:39.630 --> 04:42.750 but they have kind of self-selected 04:42.750 --> 04:46.650 on this type of this aesthetic. 04:46.650 --> 04:47.940 They've also done really well 04:47.940 --> 04:50.520 with, like, building a community 04:50.520 --> 04:52.110 and kind of engaging that community. 04:52.110 --> 04:54.600 So, you know, I predict it's gonna be here to stay 04:54.600 --> 04:57.960 for a long time, and really helpful for you 04:57.960 --> 04:59.343 to learn how to use it.