How To Install ComfyUI and Run SDXL on Low GPUs

struggling with a low GPU maybe a laptop GPU with Just 4 or 6 GB of vram and tired of waiting for sdxl models to generate an automatic 1111 well today's video is tailor made for you comfyi is your ultimate solution with it you'll create top quality images faster than ever before not only that you can use a refiner and upscaler with sdxl models and here's the kicker you can even outperform a GPU with 12 GB of vram running automatic 1111 in this video we're going to do these test together using a gtx1660 and RTX 3060 but that's not all I'll guide you step by step on how to install and make the most of comfy UI you'll learn how to load workflows that will take your creativity to the next level all without having to invest in an expensive GPU if you're not subscribed please hit that subscribe button and show us some love by Smashing the like button it helps us beat the algorithm ready let's dive [Music] in the absolute first thing we need is something called git and basically what this does it pulls files from GitHub which is sort of a main site for a lot of stuff for stable diffusion and we'll be using this in this video and future videos to come it's very easy to install so don't worry about it click on the link and when you get to this page you want to download the version that pertains to you for most of you it'll be this one so click on that and save it to your downloads folder in your downloads folder you'll see this icon here all you have to do is double click it you'll get a notification if you want to install it just click on yes and then when you see this window just simply click on install then it's going to run through the installation process that's all you really have to do next we're going to install comfy UI so head on over to this page and scroll down until you see this section here installing for Windows click on this link to download it and also if you don't already have szip click on that link to download it this will help extract comfy UI depending on your system if it's 64 or 32bit most likely b64 you want to download this version once the file is downloaded to your downloads folder simply right click click on open with szip file manager select the file and click on extract at this point you want to select where you want to install it so you can click on three buttons here and pick a drive of your choice you could even create a new folder if you want this should only take a few seconds depending on your system next you want to go into that directory in my case it was my C drive you'll see a folder here called comfy Windows portable double click on that this is your main comfy UI folder that will hold all your other assets we'll come back to this in a second back to the installation page you want to scroll down just a bit here and by the way if you're an AMD user you can run it only on a Linux environment you want to follow these steps here for NVIDIA users you want to highlight all of this line here and click on contrl C back to the comfy UI folder so you want to double click in the main comfy UI folder folder highlight the URL and put in CMD once again we'll be in the command window and we want to paste that information and click enter this is going to install X forers that helps speed up the generation time as well as P torch which is needed to run comfy UI leave the window open we'll go back to the main instructions here highlight this line here and simply paste it in the command window now if you're like me and you use other platforms like automatic 1111 how you can redirect your models confi is pretty simple so first go to your stable diffusion models folder if you're using automatic 1111 it should look something like this you would just have to rightclick copy address to text then you'll have to go to your main comfy UI folder then look for this file called extra model path yaml example you can open it with something like notepad I've got notepad++ and within this file it tells you what to do so for your checkpoint models you can paste your directory here so we're going to right click paste and then you would do the same thing for all your other files like your vaes your lauras your upscale models whatever the case may be you can paste your links here and then if you look at the top here it says rename this to extra _ modore path. yabo then on the top right you want to click on save so you can right click select rename or F2 and just remove the example from here in the period hit enter click on yes yes now if you want a custom model the place to go is civit AI again I'll leave a link in the description below but any of these top ones are great to start with now we can finally test it out and see if it's running properly when it opens up you'll see the default interface now you could zoom in with a scroll wheel or control plus and minus and you see the checkpoint here the model is already loaded let's open that up a bit and to navigate you just have to click on the workspace and move around it has a default prompt already so we're just going going to run this and to generate the image we're going to hit Q prompt here and you'll see as it passes through the process and there's our image so congratulations you've installed comfy UI if you happen to be new to stable diffusion basically comfy UI is like looking under the hood of all those other platforms like automatic 1111 invoke gii or even some of the online ones that you might be using now this is a very basic setup but the way it works is that you load your model here and that model contains all the data that will create your image in this section we see clip text encoders for our positive and negative prompts this is how we tell the AI what image we want and the text encoding is really just converting data into numbers and then at the bottom here you see something called empti latent image so think of a blank canvas so depending on what you type in here it's going to pull information from the model it'll go through something called a diffusion process where it creates all this noise and there'll be layers upon layers of that noise then it'll go through this sampler process and this is what helps to do the D noising to create your image the vad code here is sort of like the final steps and it'll convert all this information back into pixels to create your final image I definitely want to go deeper on this process in future videos but that's basically how it works before we get into some other housekeeping stuff I do want to show you how to install the plugins manager that we're going to use quite a bit in future videos once again click on the link in the description below and you'll come to this page as we scroll down here we'll come across the installation section and the method we're going to use is the second method here so the first step is to install git we've already done that the second step is to download this into the comfy UI Windows portable directory now instead of clicking on it we're going to right click save link as and we have to find this directory here comi Windows portable and save it into that folder head on over to that same folder and now you'll see this icon here install manager for portable version we're going to double click and it'll do the installation and then we're going to restart comfy UI when you restart comfy UI now you're going to see this manager option the standout feature of comfy UI is the capability to download a custom JavaScript workflow file created by users and seamlessly load it into your UI we're going to explore a workflow that promises to generate sdxl images Swift even with low vram graphics cards first we'll add some parameters in our run Nvidia GPU dobat file using a text editor next we'll download the sdxl refiner from hugging face and put it in comfu eyes models folder inside the checkpoints directory or if you are using automatic 1111 models directory save it there after that we are goinging to download sdxl vae and put it in the models folder inside the vae directory we'll need also two upscaler you can grab them directly from the civetti page and place them in the upscale models folder lastly get the following workflows V2 V2 with Lura quality and V2 with Lura fast extract these workflows from the zip files and put them inside a new folder in your comfy UI now it's time to switch to a lower-end GPU to test this workflow effectively right now I'm using an NVIDIA GTX 1660 with 6 GB of V r as you can see in my task manager here now let's open comfy UI and let's import the first workflow made by the civi user Pista first we'll create an image using the default workflow that appears when you open comfy UI keep in mind that we'll conduct these tests after generating the first image the default comy UI workflow took 2 minutes and 43 seconds to generate an sdxl image with 20 sampling steps without any Laura refining model or up scaling now let's experiment with the civit AI workflow click load and select the first Json file comfy fast low Ram V2 this may appear a bit complicated but it's easy to understand and don't worry I will do more comfy UI tutorials in future to begin load the vae we previously downloaded and for the base model choose sdxl next load the refiner we got from hugging face and select one of the upscaler we added earlier we'll stick with the default prompt in this test and then simply click Q prompt with the civid AI workflow it took 2 minutes and 7 Seconds to generate an image using the sdxl model a refiner and an upscaler now let's compare this with the RTX 3060 using the same prompts and settings with a 12 GB graphics card the image was generated in just 22 seconds this is remarkable because with the same settings using an sdxl model along with a refiner and upscaling the image in automatic 1111 would take much longer as expected automatic 1111 took 3 minutes and 50 seconds to generate one image on a 12 GB graphics card which is significantly longer than the GTX 1660 with only 6 GB of vram this is one of the reasons why I find comfy UI to be superior in many aspects and you're all set for today's video if you're an automatic 1111 user I hope this has been a helpful introduction to comfy UI I also wish that this workflow will speed up your image generation process and save you valuable time if you encounter any issues please don't hesit to let me know in the comments below also feel free to check out my previous video if you want to learn more about making money using [Music] AI

Channel: Aiconomist

Views: 4,301

Keywords: comfyUI, stable diffusion, comfyui tutorial, stable diffusion tutorial, comfyui nodes, comfyui workflow, comfy tutorial, midjourney alternative, automatic1111 alternative, sdxl comfyui, comfyui sdxl, stable diffusion comfyui, comfyui guide, comfyui install, comfyui easy workflow, comfyui efficiency nodes, comfy ui, generative ai, stable diffusion alternative, speed test

Id: tQK3DLJzPz0

Length: 11min 16sec (676 seconds)

Published: Sat Oct 14 2023