WEBVTT 00:01.110 --> 00:01.943 -: All right. 00:01.943 --> 00:04.470 What are Midjourney's capabilities and limitations? 00:04.470 --> 00:06.300 What's good about it and what's bad? 00:06.300 --> 00:09.030 One cool thing is that it's built into Discord 00:09.030 --> 00:10.560 into the community tool. 00:10.560 --> 00:13.050 Similar to Slack. You type in your prompt, 00:13.050 --> 00:14.430 it's /imagine, 00:14.430 --> 00:16.260 and then whatever your prompt is. 00:16.260 --> 00:18.480 And that makes it easier in some ways. 00:18.480 --> 00:20.580 Like you can use it on mobile pretty easily. 00:20.580 --> 00:22.290 And you can also click to upscale 00:22.290 --> 00:24.720 or click on the variance as well. 00:24.720 --> 00:27.630 It's community first because it's built into Discord. 00:27.630 --> 00:31.560 Every single member, you know, of every single user 00:31.560 --> 00:34.890 of Midjourney is a member of this platform. 00:34.890 --> 00:36.630 And because, you know, 00:36.630 --> 00:40.170 because you get a lot of people posting in there, you get 00:40.170 --> 00:42.120 to see a lot of variation in terms of like 00:42.120 --> 00:44.310 what prompts are working for different people. 00:44.310 --> 00:45.540 And you can just take those prompts 00:45.540 --> 00:46.950 and then copy them as well. 00:46.950 --> 00:49.113 So it's a real hotbed of innovation. 00:50.310 --> 00:52.830 One cool thing you can do is you can add a base image 00:52.830 --> 00:53.663 into the prompt. 00:53.663 --> 00:56.280 So you can provide a kind of an example, 00:56.280 --> 01:00.570 and that tends to really improve performance of the prompts. 01:00.570 --> 01:03.630 You can also do image blending, so this is kind of unique. 01:03.630 --> 01:05.760 You can just give it two different images 01:05.760 --> 01:07.320 and blend them together. 01:07.320 --> 01:09.690 I found that that can be quite powerful. 01:09.690 --> 01:11.910 Negative prompting is really key. 01:11.910 --> 01:13.260 It's something you can't do in DALL-E, 01:13.260 --> 01:15.270 but you can do, in Stable Diffusion, 01:15.270 --> 01:16.950 you can remove different elements. 01:16.950 --> 01:19.500 So in this example, we've got Peppa Pig, 01:19.500 --> 01:20.430 a children's cartoon, 01:20.430 --> 01:22.200 but we've stripped 01:22.200 --> 01:23.880 out the cartoon aspect 01:23.880 --> 01:26.313 and got this kind of nice fluffy toy instead. 01:27.600 --> 01:30.480 Here are weighted terms, so this is, you know, kind 01:30.480 --> 01:34.290 of like a bit more fine-grained, you know, control 01:34.290 --> 01:35.970 over the prompt. 01:35.970 --> 01:37.290 Similar to negative prompts. 01:37.290 --> 01:38.460 Like if you put minus one, 01:38.460 --> 01:41.220 then that'd be the same as negative. 01:41.220 --> 01:44.370 But you know, here we blended a little bit of Van Gogh 01:44.370 --> 01:45.330 and the, and you know, 01:45.330 --> 01:49.320 and some Picasso to kind of get the ultimate style 01:49.320 --> 01:51.960 that we wanted, which is kind of interesting. 01:51.960 --> 01:54.090 And then we also have a lot of model settings. 01:54.090 --> 01:57.300 So you know, you can choose which version you're using. 01:57.300 --> 01:59.520 So if you wanna scale back to an older version, 01:59.520 --> 02:00.690 they're all very different. 02:00.690 --> 02:02.760 Like v5.0 is a lot more realistic, 02:02.760 --> 02:04.770 v4.0 is a lot more cinematic, 02:04.770 --> 02:08.460 and they have a few different other styles and modes, 02:08.460 --> 02:10.380 but it does have some limitations. 02:10.380 --> 02:13.860 So one thing that, I guess it could be a positive 02:13.860 --> 02:15.390 or a negative is that it tends 02:15.390 --> 02:17.160 to have this fantasy aesthetic. 02:17.160 --> 02:21.240 So everything comes out, you know, relatively like fairytale 02:21.240 --> 02:22.740 or video gamey. 02:22.740 --> 02:25.530 And I think that that's likely to stay just 02:25.530 --> 02:30.510 because the community itself rates the images 02:30.510 --> 02:33.030 and the community, I think the early community were a lot 02:33.030 --> 02:35.130 of like digital artists 02:35.130 --> 02:36.600 and game designers, things like that. 02:36.600 --> 02:38.430 So it's kind of had that look 02:38.430 --> 02:39.420 and feel, it looks like something 02:39.420 --> 02:41.733 from Deviant Art or Art Station. 02:42.720 --> 02:45.780 The other thing that it doesn't do very well still is kind 02:45.780 --> 02:47.820 of like faces and hands. 02:47.820 --> 02:49.200 In some cases, like if you ask 02:49.200 --> 02:52.260 for a model, it always looks too polished and perfect. 02:52.260 --> 02:54.540 The hands themselves tend to be pretty bad, 02:54.540 --> 02:58.200 although that is better in v5.0. 02:58.200 --> 03:00.510 You know, this is like a pretty interesting image, 03:00.510 --> 03:02.280 but you know, you can tend 03:02.280 --> 03:03.870 to see the faces look really scary, 03:03.870 --> 03:06.150 not, you know, not particularly realistic. 03:06.150 --> 03:07.980 Occasionally you just have the, 03:07.980 --> 03:10.170 like a really odd thing happening like, you know, 03:10.170 --> 03:12.060 this woman's weird face here, 03:12.060 --> 03:14.100 or you know, this woman like facing the wrong 03:14.100 --> 03:15.870 way from the MacBook. 03:15.870 --> 03:18.660 So yeah, still some issues. 03:18.660 --> 03:20.850 There's no API available unfortunately, 03:20.850 --> 03:23.610 so this is really only a consumer tool. 03:23.610 --> 03:24.810 You can't really use it 03:24.810 --> 03:28.500 to, you know, use it programmatically for anything. 03:28.500 --> 03:29.820 But other than that, 03:29.820 --> 03:33.660 like I think Midjourney is like a very good tool to use. 03:33.660 --> 03:36.090 It's surprising actually how well it's done 03:36.090 --> 03:37.800 considering it doesn't have the funding 03:37.800 --> 03:40.740 of OpenAI or, you know, the open source. 03:40.740 --> 03:42.150 It doesn't benefit from open source 03:42.150 --> 03:44.970 like Stability does Stable Diffusion. 03:44.970 --> 03:48.600 So yeah, I think this is, it's carved out really niche. 03:48.600 --> 03:50.880 I think the community has driven it forward 03:50.880 --> 03:53.530 and we'll continue to drive it forward in the future.