Using Stable Diffusion on GIMP with Intel Arc Graphics

hey Bob Duffy from Intel here and I want to show you something that's pretty cool and this is running stable diffusion inside an application called and it's powered by openvino which supports Intel Arc um so if you go to the Intel Insider's Discord I've got this all well explained for you I can also show you the GitHub repository of where you'd get this but first if you want to do it go to discord.gg forward slash Intel that's the Intel insiders Discord within there we've got our creator Channel and I've got a discussion on this for stable diffusion in in here the first thing is you can go to the website there it also explains everything there are instructions to install it but what I've done here is I've added a few more instructions than they've provided here this is written a little bit more from a developer's perspective so if you're not a developer you probably should uh have um get installed if you didn't and pytorch needs to be installed so there's some very simple instructions on how to get those tools installed so that you can run this and then you'll get those instructions to set like I said on the Intel Insider Insider's Discord here um but um it's very cool if you're not familiar with is like Photoshop but it's open source so it's an image editing application so this will allow you to generate and edit images using AI inside an image editor which you could then take that and do more with but you've got things like segmented segmentation supported which will give you masking of foreground versus background objects you've got super resolution so if you want to upscale the resolution of of a grainy photo you can do that you can do style transfer so you can you know add different looks to images you can do in painting which is something that's very cool with stable diffusion so if you want to you know within one of your images you want to be able to change or edit out something or add in sunglasses on top of a person things like that you can do in painting which that will do it just fine and it's pretty cool um and then you've got what we typically think of stable diffusion for which is essentially using text to describe an image and then have that image generated for you um I've got examples of this um in The Insider's Discord here here's an example of a portrait of a robot from the 1940s horror film taken with a 50 millimeter lens at a film studio Sound Stage and this is what it comes up with which is it's frightening and cool at the same time so if you're trying to create an interesting looking robot and you just weren't inspired so maybe you're doing this for a video game um or maybe it's concept art for a film yeah use stable diffusion to help you on the creative process also you can use it for inter interior design if you're trying to come up with a color palette or you're trying to come up with a scheme for furniture or a design or drapery or textures that could be in a room you know describe what it is that you're looking for and have a stable diffusion come up with those ideas for cosplay or character design of a particular kind of period character it'll do that as well and just prompt away and do as much as you want because the cool thing about stable diffusion when you run it locally on your computer this is free you're not having to pay for a subscription for it you can you can essentially keep going at it and because it supports Intel Arc and gpus it's pretty quick it's it's Snappy to say the least so I'm going to give you a rundown of this we're going to give this a shot so when it's working you go to the layer menu you select open Vino we will do a stable diffusion prompt when you've got that you can select CPU or GPU we'll kick it off with GPU and what I do here is let me find my notepad somewhere here hold on see where I got it all right your Notepad there you are all right um I went ahead and I I typed in some prompts here and uh so with this version of stable diffusion it doesn't persist The Prompt um so you're going to want to kind of keep your prompt somewhere so if you want to make an edit or change to that prompt um just keep that somewhere so just notepad works so I'm going to work with notepad so again let's go here Vino stable diffusion and we'll put this here uh let me move this out of the way I'm going to go ahead and clear this timer and have to remember every time I do this is to do that so we'll uh run inference we'll start the timer and what I notice is um before it starts inferencing there's about 20 to 25 seconds or so of of something going on um and and we'll see this on gpus our CPU as well we'll try both of them and then once it gets past this initial say 25 seconds uh then it starts the inferencing and that happens really quickly especially on gpus so we should see that just about to kick off okay we're started the inferencing here okay and here we are um all right so this is it so um that looks pretty good to me that's uh nice in interior design so we've got an interesting color palette I think I said use a neutral color palette it did that interesting that it shows black for the walls which is a super interesting look and so again if you're into interior design you're doing some decorating this may be a way to get that done ah I forgot to look at what time that was so you guys maybe roll back in the video and see when that happened what time we were at so we will do this exact same prompt but I think it was right around a little over 30 seconds we'll do this I think it was maybe 25 seconds is where um it moved from device to inferencing and then it took a handful of seconds after that in order to get it going so we'll check this out on CPU let's go ahead and do the same thing stable diffusion we'll paste in our prompt but we will say CPU instead so let's go ahead and run that inference and um it's going to create a new image every time it does this so it'll keep all of your images together there um let me let me redo this and cancel let's see if I can cancel okay I'll let this go and we'll we'll try it again I didn't do the timer but what we can see here is where this was done with whatever this is to 7.1 this is going to count up quite a bit so I'm thinking this is this is running you know about 10 seconds longer than the GPU and we can run it again to be sure okay so let's rerun that I will hit play this time now it seems to like it yeah oh one thing I should tell you is whether you're running CPU or GPU it's not going to change the quality or the content that's in that image so if you like an image better when it was done on CPU that doesn't mean that that was because it was done on CPU that's just essentially a different seed a different diffusion that it was taking and that's that's defining what's in the image as well as your prompts um so it really the the CPU versus GPU is really just going to determine the time so that this going like I said this does seem to be a bit longer than it was for GPU I think we were maybe 30 something seconds here and stop okay so 55 seconds here again so we have these three images the first one was done by GPU these two were done by CPU so the image image quality you know the kinds of stuff you're getting different looks that's not dependent upon CPU or GPU but essentially that the amount of time that it takes uh uh certainly so let's let's try another prompt here and we'll go back to GPU so we'll try a portrait of a female mid-20s and Renaissance clothes long red hair shot with a 50 millimeter lens Studio lighting so stable diffusion go back to GPU we're gonna we're gonna hit run inference we're going to clear and go um okay so um now let me explain the prompt a little bit because I I think prompting is something it's a bit of an art that people start to learn over time notice I said a portrait of a female I didn't say a photo of a female I said a portrait of a female when you use the word portrait it tends to kind of focus in on the person when you say photo of sometimes it can actually put a photograph of that person in your image okay so here we go um so that didn't take long at all um so in comparison so yeah we're in the low 30s to mid 50s versus uh GPU versus uh CPU on this so here we go uh a nice picture here uh this is a similar picture that I did I think the difference here is I said Renaissance clothing on this one this one I did not so and again the prompting is the key thing here portrait seems to focus in on a person if you say photo you might end up with actually something that looks like a physical photo in the scene if you put in the age you'll get that something like you know a hair color or eye color or shape of face if you want to put that in describe the type of clothing especially if you want it to be a period piece put in a year for that type of clothing if you're looking for that often if you're saying something is happening maybe in the 20s or the 30s your image coming back will be black and white so definitely if you wanted it to be color you could say a color portrait of a female in the mid-20s say from 1935 you could say something like that long red hair shot with and then describe the lining or the scene or the environment that should be in that photo when you say something like shot with and when if I didn't have this information in here if I said a color portrait of the female mid-20s 1935 long run here it might be stylized in something that's a painting or a rendering and so if I wanted to look more like a photograph I I describe the the photography being used if you want we can give that a shot um layer if you don't plug-in stable diffusion paste that in and run that inference and see how we do um so these are some other examples that I got on the living room to prompt so I think in this one I asked it to be fancy and I gave it a color palette which is pretty cool let's see what we end up getting here okay it should be uh kicking off here in a half moment oh interesting yeah so I didn't describe the clothes so focused in on the person right from 1935. so it'd be interesting so this is where you started get into a prompt crafting right so again if you just if you copy your prompts in and paste them in then I can go ahead and edit this a color portrait of female [Music] um from I could say a portrait of a 25 year old female taken in 1935. long red hair decorative clothes shot with a 50 millimeter lens um interior Studio let's give that a shot okay and paste that in run the in first we'll skip the time I think we got a good idea of the difference between CPU GPU in terms of how this is calculating um but it's really nice to see it run on Arc what we can do here is look at performance monitor here and what's going on not sure if this would be a compute calculation or 3D calculation so if I look at compute we see a little bit here it tends to be a little bit more spiky but definitely there's some 3D going on and my other GPU it really is nothing going on so while I have two gpus not much is happening there and let's see what we have ah look at that okay so when we describe the close it definitely changed the photo so now we have a photo where we're seeing the upper torso sort of thing now there's a little bit too much red going on so you could adjust uh your prompting there uh you know you could say neutral colored clothes Etc um so so there you have it this is stable diffusion running inside of powered by open Vino which allows you to Target various architectures CPUs gpus and such including Intel Arc so if you do this please post results questions and discussion on the Intel insiders Discord that's discord.gg forward slash Intel I'd love to see what you're doing and love to hear your feedback please provide it I'll do my best to get it to the Intel open Vino team so thanks for watching and until next time see you on Discord

Channel: Intel Graphics

Views: 16,028

Keywords: Intel, Intel Graphics, Graphics, Intel Corp, Discrete Graphics, Graphics Hardware, Media, Media Hardware, dGPU, GPU, Intel Arc, Intel Arc A-Series, A-Series, XMX, Ray Tracing, Intel Arc Graphics, Hyper Encode, AV1, XeHPG, Community, PC, Video, PC Parts, Technology, Computers, PC Gaming, Video Games, Graphics Card, Preview, Performance, Blender, Rendering, oneAPI, GIMP, Stable Diffusion, AI Art, AI, Artificial Intelligence, Intel Core Processor, Intel Core, Gaming, Specs, Computer, OpenVINO

Id: q8xJjPBjqso

Length: 16min 0sec (960 seconds)

Published: Thu Apr 06 2023