WEBVTT 00:00.027 --> 00:01.740 -: Hey, I'm gonna walk you through 00:01.740 --> 00:04.290 how to get stable diffusion web UI running 00:04.290 --> 00:07.170 or otherwise known as AUTOMATIC1111. 00:07.170 --> 00:08.370 This is a way 00:08.370 --> 00:11.340 to get stable diffusion running locally on your computer. 00:11.340 --> 00:12.750 And before you do this, 00:12.750 --> 00:14.730 you will need some technical knowledge. 00:14.730 --> 00:17.160 It's not straightforward, it's a bit complicated. 00:17.160 --> 00:18.930 There's a lot of complicated parameters. 00:18.930 --> 00:20.670 You might not know how they work. 00:20.670 --> 00:23.940 There's also a requirement which is you need a GPU. 00:23.940 --> 00:25.803 So I'm running this on a Mac M2. 00:26.760 --> 00:28.710 There's also, you know, windows version 00:28.710 --> 00:33.300 for NVIDIA style GPUs if you have a gaming pc. 00:33.300 --> 00:35.400 But that's something that needs to happen 00:35.400 --> 00:38.700 'cause it can't, it's pretty heavy load 00:38.700 --> 00:42.330 and it can't really run just on your normal computer. 00:42.330 --> 00:45.690 You need a graphics processing unit, a GPU. Cool. 00:45.690 --> 00:48.450 This is Stable Diffusion web UI. 00:48.450 --> 00:50.160 Just wanna show you what it looks like when it's running 00:50.160 --> 00:52.530 and then we'll talk about how to install it. 00:52.530 --> 00:55.710 I'm in the stable diffusion web UI folder 00:55.710 --> 01:00.710 and then I just say bash webui.sh and it's running 01:00.875 --> 01:03.510 and it's gonna install anything it needs 01:03.510 --> 01:07.140 and get it to the place where it's actually running. 01:07.140 --> 01:09.000 You might get some error messages here, 01:09.000 --> 01:10.530 which you'll have to figure out. (laughs) 01:10.530 --> 01:12.390 Sometimes it's a bit complicated, like I said, 01:12.390 --> 01:14.610 there's a technical ability to it. 01:14.610 --> 01:16.920 It says it's just installing the requirements 01:16.920 --> 01:20.730 and once it's running, it will be available on this URL, 01:20.730 --> 01:23.520 which you'll be able to grab from the terminal. 01:23.520 --> 01:26.820 So I have a few extensions installed to try, 01:26.820 --> 01:28.870 so I am bringing some of these things in. 01:29.717 --> 01:33.947 Okay, yeah, here we go. This is the URL. 01:33.947 --> 01:37.203 So you can see there, paste it. 01:39.030 --> 01:42.120 All right, we have it so we can change models up here. 01:42.120 --> 01:47.120 I'm using the standard stable diffusion, the v1.5, 01:47.310 --> 01:50.760 a lot of people use version 1.5 over the version two 01:50.760 --> 01:53.580 or plus models just because it's a bit more flexible. 01:53.580 --> 01:57.210 There's a lot of user generated type content. 01:57.210 --> 01:58.950 People have trained models based on this, 01:58.950 --> 02:01.170 that are quite useful and all of those, 02:01.170 --> 02:03.570 like when you download them, they show up here. 02:03.570 --> 02:04.920 I'll give you a quick example. 02:04.920 --> 02:09.160 So just say like a cat wearing a cape 02:10.260 --> 02:12.930 and hit generate and this is where you put the prompt. 02:12.930 --> 02:15.690 You can also do negative prompts in here. 02:15.690 --> 02:18.930 There's, you know, a lot of functionality 02:18.930 --> 02:21.750 in terms of the parameters and stuff you can change. 02:21.750 --> 02:23.610 Don't worry about that too much right now. 02:23.610 --> 02:25.470 We'll cover that. So here we go. 02:25.470 --> 02:27.210 There's a cat generated with a cape 02:27.210 --> 02:29.190 and then I'm just gonna show you how 02:29.190 --> 02:31.554 that looks like a different model. 02:31.554 --> 02:35.610 So this is a cat wearing a cape. 02:35.610 --> 02:36.900 And actually by the way, 02:36.900 --> 02:39.240 one thing that's interesting is the every image 02:39.240 --> 02:40.080 that's generated, 02:40.080 --> 02:42.920 the prompt is saved as metadata in the image. 02:42.920 --> 02:44.400 So you can also see the seed 02:44.400 --> 02:47.400 and what model you used, et cetera, what sampler. 02:47.400 --> 02:49.050 That's a cat wearing a cape. 02:49.050 --> 02:49.920 I'm just gonna show you, 02:49.920 --> 02:52.713 I'm just gonna try the inkpunk-diffusion model. 02:57.390 --> 03:00.810 And that's working. And if you are downloading a model, 03:00.810 --> 03:01.950 lemme just get generate here. 03:01.950 --> 03:04.740 If you're downloading a new model, say from civitai, 03:04.740 --> 03:07.260 bear in mind that some of these are not safe for work, 03:07.260 --> 03:10.230 but if you download the model then you need 03:10.230 --> 03:13.650 to put it into your folder here. 03:13.650 --> 03:16.380 So this is stable diffusion web UI 03:16.380 --> 03:19.389 and then there's this models folder 03:19.389 --> 03:21.840 and then you have stable diffusion 03:21.840 --> 03:22.950 and this is where you stick it 03:22.950 --> 03:24.480 and when it's there, 03:24.480 --> 03:28.470 then you can just hit this refresh and it should show up 03:28.470 --> 03:29.940 or you can restart. 03:29.940 --> 03:33.210 Yeah, here we go. So this inkpunk-diffusion 03:33.210 --> 03:35.340 it didn't do anything different really 03:35.340 --> 03:37.590 'cause I didn't use the trigger word, 03:37.590 --> 03:40.803 but specifically there's a trigger word nvinkpunk. 03:41.774 --> 03:45.180 If I hit generate, it's gonna do it in this style. 03:45.180 --> 03:47.970 So this is a dream booth model that we can use. 03:47.970 --> 03:50.280 But there are, you know, lots of different types. 03:50.280 --> 03:52.620 Some have trigger words, some don't. 03:52.620 --> 03:54.063 You can see how that works. 03:57.540 --> 03:59.940 Yeah, here we go. So now it's in the ink punk style 03:59.940 --> 04:02.343 where you can see it's diffusing towards that. 04:10.078 --> 04:14.100 There we go, an inkpunk cat in a cape looks a lot cooler. 04:14.100 --> 04:15.510 So you can change the sampling steps. 04:15.510 --> 04:19.050 This is how many, yeah, how long to run diffusion process. 04:19.050 --> 04:20.670 You can also change the sampler. 04:20.670 --> 04:23.580 I just used Euler a, ancestral that stands for, 04:23.580 --> 04:25.770 there's a lot of differences between them, 04:25.770 --> 04:27.300 but the ones that stand out, 04:27.300 --> 04:28.830 a lot of people use eular a, 04:28.830 --> 04:33.090 a lot of people use DPM++ 2M Karras, 04:33.090 --> 04:37.020 this one here and then DDIM. 04:37.020 --> 04:38.640 That was like the original one 04:38.640 --> 04:39.930 that was like made specifically 04:39.930 --> 04:42.060 for stable diffusion, I think. 04:42.060 --> 04:43.710 This is how many images you get. 04:43.710 --> 04:46.200 So this would generate four images in a row 04:46.200 --> 04:50.310 and then this batch size is like how many images per batch 04:50.310 --> 04:51.660 that can be multiplied together. 04:51.660 --> 04:53.727 If you don't have a lot of VRAM 04:53.727 --> 04:56.430 and then your batch size, you wanna keep that small 04:56.430 --> 04:58.950 because it's generating at the same time 04:58.950 --> 05:01.363 it can really lag on your computer. 05:01.363 --> 05:06.180 A CFG scale is how much of a difference the prompt makes. 05:06.180 --> 05:09.270 So how close it is to the prompt versus how creative. 05:09.270 --> 05:13.397 So it's similar to temperature on the GPT-3 05:13.397 --> 05:15.060 or GPT-4 interface. 05:15.060 --> 05:16.980 You can add extensions here. 05:16.980 --> 05:20.220 So I've added control net, which I'm not gonna walk in 05:20.220 --> 05:21.450 but walk you through. 05:21.450 --> 05:24.840 But there's a lot of like cool stuff you can add in here 05:24.840 --> 05:26.310 and they'll show up in different places 05:26.310 --> 05:28.170 in the user interface. 05:28.170 --> 05:31.118 The way that you add extensions is if you go here, oh sorry, 05:31.118 --> 05:34.523 extensions and then available, load from, 05:34.523 --> 05:38.730 and it loads it all from this, this JSON file 05:38.730 --> 05:40.980 and you can see which ones there are and what they do. 05:40.980 --> 05:42.660 And then you can install them. 05:42.660 --> 05:45.000 You can also load them from a URL 05:45.000 --> 05:47.610 if you just find one out there on the web. 05:47.610 --> 05:49.590 But obviously be careful 'cause you're loading code. 05:49.590 --> 05:52.260 Some of them show up as their own separate tabs. 05:52.260 --> 05:54.150 Inpaint Anything is one example. 05:54.150 --> 05:57.420 They have, in some cases, they have like settings here 05:57.420 --> 05:58.560 that you can change. 05:58.560 --> 06:00.480 There's also like a lot of built-in extensions, 06:00.480 --> 06:03.990 like the image to image stuff you have inpaint in here, 06:03.990 --> 06:04.920 which I won't go through. 06:04.920 --> 06:07.080 You have interrogate clip, 06:07.080 --> 06:09.480 so you can reverse engineer the prompt from an image, 06:09.480 --> 06:10.980 which is pretty cool. 06:10.980 --> 06:13.800 There's a lot of parameters here you can mess around with. 06:13.800 --> 06:18.090 There's also the different scripts you can run. 06:18.090 --> 06:20.370 So like the X, Y, Z plot would generate, 06:20.370 --> 06:22.440 you could generate like say the same image 06:22.440 --> 06:27.420 but with five different values for CFG scale for example. 06:27.420 --> 06:30.930 So there's a bunch of different things you can do here. 06:30.930 --> 06:33.960 In the extras is where you do the scaling. 06:33.960 --> 06:37.740 So if you wanna upscale the image to a higher resolution 06:37.740 --> 06:40.170 or higher size, and then this is where you would do it. 06:40.170 --> 06:42.480 In extras you can also do batch processing 06:42.480 --> 06:45.313 for multiple images if you need to. 06:45.313 --> 06:48.930 Cool. That's the user interface. 06:48.930 --> 06:50.850 And if you ever get confused 06:50.850 --> 06:52.590 or you don't know what different things do, 06:52.590 --> 06:54.870 the wiki here is really good. 06:54.870 --> 06:58.140 Like it has a pretty detailed understanding 06:58.140 --> 07:00.210 of what these different things are. 07:00.210 --> 07:03.390 You know, check on that. But to install specifically, 07:03.390 --> 07:06.750 I'll just show you what I did to install on the Mac, 07:06.750 --> 07:08.520 and this is on the Apple Silicon. 07:08.520 --> 07:11.490 So it's the M1 and two Macs in particular. 07:11.490 --> 07:12.810 I used Homebrew, right? 07:12.810 --> 07:17.810 Once you have Homebrew, you run this in the terminal. 07:18.300 --> 07:21.870 I say if I just press Ctrl+C to cancel this, 07:21.870 --> 07:26.870 if you run that, it will install the cmake and protobuf 07:26.940 --> 07:29.340 and get wget. If you don't have them 07:29.340 --> 07:33.060 and then you just do a git clone when you're in your folder 07:33.060 --> 07:36.510 and that will just pull down from GitHub 07:36.510 --> 07:38.430 all the stable diffusion web UI code 07:38.430 --> 07:40.440 and that will give you that local folder 07:40.440 --> 07:42.330 that I was showing you before. 07:42.330 --> 07:46.650 Then you need to download the stable diffusion models. 07:46.650 --> 07:49.895 It has some examples here like where you can get them from. 07:49.895 --> 07:52.830 Version 1.5 is the one that I use. 07:52.830 --> 07:55.020 There's also a special one for Inpainting, 07:55.020 --> 07:56.220 which is quite useful 07:56.220 --> 07:59.790 and or there's version two if you want as well. 07:59.790 --> 08:01.800 And typically you just click it 08:01.800 --> 08:04.410 and it will take you out to somewhere like hugging face 08:04.410 --> 08:06.720 where you can, you know, download the model. 08:06.720 --> 08:08.670 The files and versions are in here 08:08.670 --> 08:10.920 and I think basically you just need this 08:10.920 --> 08:12.900 checkpoint file here. 08:12.900 --> 08:17.520 So if you click that, it'll start downloading. Okay. 08:17.520 --> 08:21.450 Then once you have, once you have that available, 08:21.450 --> 08:23.790 you drop that in your models folder 08:23.790 --> 08:25.800 and then you can just cd like change directory 08:25.800 --> 08:30.800 into stable diffusion web UI and then run bash webui.sh 08:30.989 --> 08:34.200 or just this ./webui.sh. 08:34.200 --> 08:36.180 And that will do everything for you. 08:36.180 --> 08:38.311 Now if you're running any issues 08:38.311 --> 08:42.083 there's actually quite a lot of information here. 08:42.083 --> 08:45.150 There's like all these issues you can search through 08:45.150 --> 08:47.760 and if you just search for the error message you're getting, 08:47.760 --> 08:50.760 quite often people have already written how to solve it. 08:50.760 --> 08:52.320 There's a lot of pull requests 08:52.320 --> 08:54.990 where they're trying to improve some of the UI 08:54.990 --> 08:56.940 and some of the issues that you run into. 08:56.940 --> 08:59.150 And there's like specific discussions. 08:59.150 --> 09:04.150 So that is how to get it running on Mac. 09:04.200 --> 09:08.310 On Windows so let's assume you have a NVIDIA GPU. 09:08.310 --> 09:09.720 Again, the process is similar. 09:09.720 --> 09:13.350 They don't have homebrew on the Windows, 09:13.350 --> 09:16.410 but basically the way it works is you download this 09:16.410 --> 09:18.450 and extract the zip file 09:18.450 --> 09:21.060 and then you double click the update.bat 09:21.060 --> 09:22.020 and then that will get it 09:22.020 --> 09:23.700 to like the latest version essentially. 09:23.700 --> 09:25.350 And then there's a run.bat script. 09:25.350 --> 09:26.750 You just click that and that should launch it. 09:26.750 --> 09:29.250 It will download like all the files that you need. 09:29.250 --> 09:32.280 And then you should see this like running on URL. 09:32.280 --> 09:35.220 And it's the same URL that you get on a Mac. 09:35.220 --> 09:37.530 So that is, that should work on your computer 09:37.530 --> 09:39.060 if you have a GPU. 09:39.060 --> 09:42.150 But there's also like multiple methods if you have issues. 09:42.150 --> 09:44.940 So you can do the get clone method like you do on Mac. 09:44.940 --> 09:47.880 You know, you can also use it on Linux 09:47.880 --> 09:49.920 and there's different instructions by the way, 09:49.920 --> 09:54.420 if you have an NVIDIA GPU versus if you have an AMD GPU. 09:54.420 --> 09:57.030 So just pay attention to that. 09:57.030 --> 09:58.560 But in general, it's always the same thing. 09:58.560 --> 10:00.237 You get the code from GitHub 10:00.237 --> 10:03.634 and then run like the windows, the .bat file 10:03.634 --> 10:08.310 or the .sh file, depending on which file system you're on. 10:08.310 --> 10:10.800 All right. Hopefully that's helpful 10:10.800 --> 10:12.000 because this is open source, 10:12.000 --> 10:13.650 you can literally just have a look 10:13.650 --> 10:15.934 and see how they do different things. 10:15.934 --> 10:19.730 If you want to know how do they do the, you know, 10:19.730 --> 10:23.190 how do they input all the prompts from a file, 10:23.190 --> 10:25.547 you can just see, you know, how they've done that. 10:25.547 --> 10:28.920 That's the code here, you can make updates to this 10:28.920 --> 10:30.270 if you want just locally 10:30.270 --> 10:32.820 or you could actually push 'em back into the main repository 10:32.820 --> 10:35.010 for other people to benefit from. 10:35.010 --> 10:36.570 So yeah, lots of fun. 10:36.570 --> 10:38.700 You're gonna learn a lot about image generation 10:38.700 --> 10:40.680 when you can just generate these images for free 10:40.680 --> 10:42.810 on your own computer and just leave it running. 10:42.810 --> 10:44.640 So you know, have fun with it, 10:44.640 --> 10:46.350 check out all the different parameters, 10:46.350 --> 10:48.360 look up the definitions of what they are. 10:48.360 --> 10:50.430 And that's really the best way to learn. 10:50.430 --> 10:52.470 It's just experimentation. 10:52.470 --> 10:54.370 All right. Hopefully that was helpful.