Install Stable Diffusion on windows in one click! AMD GPU's fully supported!

hey there H labbers and Engineers Fe engineer here so today we're going to take a look at setting up stable diffusion for AMD gpus in windows with a few more functionality pieces than what automatic 1111 usually lets us do since we don't have rockem quite yet for this we're going to try out nod AI shark it's fairly similar it has a lot of pieces but it runs stable diffusion differently so I want to spend a minute talking about who this is really good for and who should use this versus who maybe should not use it so a couple of the reasons why you might want to use it first of all it is a oneclick install it is dead simple secondly works on Windows it has has fast iterations per second it allows in painting it's very easy to use this tool because it has somewhat limited options it has the newer Samplers like DPM Plus+ 2m Caris which I constantly get asked why that is not in the other direct ml web UI and of course you can download custom models from civi and things like that so some reasons you might not want to use it it is really slow to change models it is really slow to change image dimensions and things like that it downloads a lot and it has pretty limited options when you want to generate images it just does not have as many settings and parameters and things that you can fiddle with like some of the other automatic web uis so with that let's take a look look at installing shark from nod AI so the very first thing we're going to do we're going to come over to the GitHub page for shark we're going to look at this prerequisites because it says you have to do this and for AMD users it just says download the latest driver great well I imagine virtually everybody already has the latest driver after that download the stable release so let's click on the stable release and the stable release there's just this dotexe file all we have to do is just click on this and download it and it's about 475 megabytes once it downloads one of the before you ever run it make sure that you go somewhere on your computer that you want to save it make a new folder so now we have a specific folder for it and then I'm going to go over to my downloads and I am just going to copy the exe file into it once you copy the exe file then you can double click on it to get an idea of the number of things that it will be downloading you can see here in my downloads folder all of these things that it went ahead and downloaded many of them multiple gigabytes this is why you want to create your own folder to put it in and run the exe from there when you run the exe unsurprisingly it will open up a big uh command prompt and it will start downloading a bunch of things it'll run a bunch of settings it'll configure things it does it all by itself it doesn't ask you anything and thankfully it'll do things like say found device AMD Radeon RX 7900 XTX and it will continue just going along one of the reasons why this is sort of fast but sort of slow is that it uses V for the image generation which is great it's actually really fast the only problem is it effectively has to compile Vulcan shaders virtually every single time that you change models or change sizes Etc that's why I'm saying if you want to generate a lot of images that are like 512 x 512 and you just want to generate generate generate this will do great if you want to use one model and then change to a different model and then change the size this is not going to do well this is just going to take a very long time once you are up and running after Double clicking that exe watching all of the many downloads many of the many gigabytes you will end up with a screen that looks sort of like this and I can't remember what it is the generic prompt is something about a car with smoke coming out of the tires like a super but anyway so I went ahead I went to civii I downloaded a different model this one is just called real cartoon 3D I put in a prompt Super Mario highly detailed uh and I left it at 512 x 512 I am using DPM Plus+ 2m Caris it's named something differently but in this one it's it's named DPM solver multi-step Caris Plus+ so we have a lot of different Samplers here that we can play with steps out of the box it only goes up to 100 the CFG scale for some reason on here it puts it at 7.5 it does have a low vram setting it does have highres fix I would recommend don't use it because just like I told you how changing models takes a long time if you use hyra as fix it's going to generate your model this way and then load in a new model and do H res fix and so it will take a really long time uh down here in device you'll see I've got Vulcan zero Vulcan 1 Rock zero Rock one CPU the Rockham doesn't work because Rockham doesn't work in Windows CPU as slow as can be Vulcan zero is my 7900 XTX and Vulcan 1 is actually my integrated GPU which obviously we do not want to use for this but you can see that I generated this 512 image that's supposed to look kind of like Super Mario Brothers and what we'll do because this space is somewhat limited we'll send it to inpaint and then in painting we'll swap over the prompt we'll swap over copy and swap over the negative prompt and now maybe inside of here maybe I want all of this to go away whatever that was I didn't like it so let's do that so with here we can set our set our scheduler height width don't change it steps you can inpaint the whole area generate image oh what is it doing is it oh yeah it's got to change models this is going to take a few minutes and you'll see that it did in fact change and and generate this within painting exactly as we expect it left basically everything the same it only changed this one little area in the bottom corner just so that people can see now that it's all set up I'll do it again and now it changed this background and you'll see that for me it took 4.5 seconds and that's at 100 steps so it does end up getting about 20 some iterations per second sometimes a little faster sometimes a little slower for me and that is roughly comparable and actually slightly better than what I get with rock you may notice up here at the top there's something called chatbot don't use it I tried using it it downloads these seven llama 7B models that are 13 or 14 gab each and it just didn't seem to work appropriately I did eventually get it to work it just was terrible for anybody who is wanting to download a model the the one that I used today was real cartoon 3D I have not tested stable diffusion XL models running on shark but in civii you can come down to any model you want get download options click on it this one had a pruned version and a non-pruned version I used the non pruned version version once that downloaded because I ran my exe file straight out of downloads first it ended up putting my models folder inside of downloads but for you since you ran your exe in some folder inside of there you should also have a models folder you can double click on it and all you will need to do is any file that you download like from civi like a safe tensor file just drop it straight inside a mo of models and after that you will have to exit out of shark entirely and then reload shark from the exe entirely and after you do that it will pick up this model there's no way for it to pick up the model in real time with it open and running but you'll see that here inside of shark AI you'll see it has that real cartoon model and it will also tell you here the custom model path is this path right here and for you wherever that model path is that's where you need to drop your models one last point on the shark web UI right here you will notice there is not a spot for extensions meaning that effectively what they have in here in painting out painting image to image upscaler chatbot Etc what they have in here is what they have there is not really a good system in place to extend it or anything like that and for many people this may not be a big deal or a problem at all however if you wanted to for instance install dream Booth you can't if you wanted to put in control net nope it is not in here so again there are a lot of people that are going to find this to be great and it's really easy to install just literally one click but if you wanted more control and extensions and things like that this may not be the solution for you thank you so much for watching I hope this has helped some folks who uh wanted to get stable diffusion up and running but really didn't want to deal with Linux or rock or the command line very much this is in fact a oneclick installer so I have to applaud no AI for building this in the way that it does seem to just sort of work good luck thanks so much guys have a great day thank you so much for watching my videos hom labbers and Engineers I create and edit all these videos on my own so any likes and subscribes will massively help out the channel and allow me to continue creating content to help people if you got value out of this please consider liking and subscribing to the channel to be notified when new content drops if there's something I've not covered but you would like to see a video on it please leave a comment down below and again a massive thank you to everyone I hope you have a great day

Channel: FE-Engineer

Views: 16,040

Keywords: AMD, Stable Diffusion, Nod.ai, Shark, AMD GPU, AI, Windows 10, Windows 11, How to install stable diffusion in windows

Id: 4mcdTwFUxMQ

Length: 13min 15sec (795 seconds)

Published: Mon Dec 04 2023