Did Vlad Diffusion/SD.Next Just Destroy Automatic1111? Full guide.

Hello friends are you looking for the latest Technologies and advances in stable Fusion right now but feel the updates it's an automatic eleven eleven are just a little too slow right now then you need to have a look at the Vlad diffusion and the community is all over it right now among other things AMD Nvidia and Mac M1 support Torch 2 sdp for faster Generations over xformers quicker startup time and build extensions that there's a lot more let's find out together all right so first we're gonna get this installed and if you aren't interested in that you can just skip forward to when we're using the tool but you're going to use the link in the description below and you're going to get to GitHub and to Vlad mandate Fork of automatic 1111 and if you scroll down a bit you're going to see in the install here you're going to need python git and if you're on a video You're gonna need the Nvidia Cuda toolkit and then we're going to clone the repository so let's do it all in order first you're going to go to Python and make sure here that you do not I repeat do not download the latest version you can't use Python 3.11 because it doesn't support Pi torch right now so what you're going to do is you can find in the list here and I recommend using a 3.10 I usually do 3.10 seven because that's always worked for me you can probably if there's an app 0.8 or 0.9 or whatever you can probably use that just make sure it's 3.10 press the download scroll down then find or Os either Mac or Windows I have a PCS on my windows I'm going to press the windows installer here I'm downloading this start the dot exe once you get this open you need to click this little checkbox here add python to path very important and then just install now this will be a fairly quick process depending on your computer for closing that so we can close this python page now we're going to go to git so you can either Google that or check the link in the description and again check your OS I'm going to download for Windows I want the 64-bit Windows setup I'm opening that yes I wanted to make changes and I am installing now it's extracting all the files and installing git on my computer and after this we're going to continue with the Nvidia Cuda tool because I use a video card in the RTX 3080 to be specific for the close that one the windows one I'm just going to pick the 11 Here Local one now this file is fairly big it's a 3 gigabyte file if you have problems downloading this you can use the network installer then you have a smaller file and it will install in bits the bits you need then you're going to open that exe just press OK here and now the installer is extracting on your computer and as soon as this is completed the toolkit installation will start and again it's a fairly easy process just takes a little bit of time this one the green continue we're just gonna run the express recommended here now it says here for me no supported version of Visual Studio was found but we're not going to use any of that so we're just going to check the box and press next so it is now finished and don't be afraid if your screen flickers that's gonna be completely normal at least during the installation not after it I'm gonna de-click those and then just close now we're gonna go back here and there's a little snippet of code here which is this git clone and then the link there's just gonna copy paste that will first copy then you're just going to go into your folder where you want it and this is my users folder I'm gonna go up here I'm going to type CMD that will open a window for me here so this is where I will paste it and then press enter now if you have automatic 1111 already installed this might be an error that you get fatal destination path automatic already exists and is not an empty directory because it uses the same folder so we're gonna do is we're gonna create a new folder here let's call it Vlad and now you can either do the same thing go CMD here or if you're in the previous you can go CD which is change directory either way is fine so now it's downloading all the files from GitHub and now that is finished you can go inside your folder and here are all the files and it's very similar to automatic 1111 because it is a fork of that and like it says here it's going to be the same thing run desired startup script to install dependencies and extensions of start server which is the web UI in automatic 1111 it's called Web UI user who is used with UI so we're going to run this batch file here and this will be the main part of the actual installation now the git clone that was just to copy the files or download the files from GitHub to your computer this will be probably the lengthiest process of the installation once that's finished you're going to see something like this and this took quite a while for me and it looked like it had stopped several times but it didn't so just stick with it probably took like 20 minutes on my fairly fast computer anyway you're gonna copy paste this open a browser window paste that in then you're gonna be inside Vlad diffusion and now this is fairly similar to automatic 11 11. it's a it's a little different style there are themes to choose from the base is kind of the same so you have from text here which you which is your text to image you have from image which is your image to image the image info is the PNG info tab you can drop your image here and get the info from previously rendered image the process here that's what used to be the extra so you can upscale do a batch process stuff like that let's just do a quick render and see how this works I'm going to load a style here and you can find these in the video description there's a link in the these are all free to download you can select multiple here if you want the positives or for anyone I'm just gonna use these two for now I'm gonna put beautiful Nordic forest with the Fjord I'm gonna have the depth to 25 at the Euler a I'm just gonna run a regular old image here nothing special there just a regular old image as you can see here you have control net already installed you don't need to install that manually so that's great especially for any beginners out there you don't need to um well you don't need to manually install it but check the extensions here the inbuilt extensions are you have the Loris you have the the ldsr the skewitness winr that those are for upscaling mainly the like core is is installed you have a clip interrogator and multi-fusion upscaler you had controller that we talked about I thought we're gonna have dream Booth I was under the impression that we are that doesn't seem to be the case let's check the available extensions here so we have shivam's dream Booth here however it doesn't seem to be all installed so we're just gonna install that and if you scroll up you can see it's processing here if you wonder what's going on you can also see inside your now if you're brand new to stable fusion and generative AI I'm just going to give you a quick overview what this is here is we're gonna create your image and we just created a Nordic Fjord here so this will be where you put in what you want let's say we have man on ship here but now there are a lot of settings that you can choose from the sampling method is basically the tool that creates your image you can mostly leave this at default for now I usually use Euler a at a sampling step of 25 but because I am I've been doing that for a long time it's fast and it works well now if I do nothing at all and generate I'm just gonna get an image of this now we didn't get a man on the ship but at least I got a ship you might need to create more batches to get your desired results anyway the sampling steps is how many steps if you see it as a frames to generate your image so it will take 25 frames to 25 steps to reach this image this image just starts as noise and then you can see that when you generate that it starts as noise and then it takes well it's too fast for you to see each step makes the image a little bit better each time and width and height here is basically just the size of the image the default is 5 12 by 5 12 for most models there are some that do 768 as well but if you don't have a specific model for that this is what you're going to use the batch account and the batch size that's how many images or how many batches you're gonna create and the CFG scale here that that's how much the air will listen to your prompt up here so a high scale Force The Prompt onto the image and the low scale will let the AI be more creative so if you set this to zero we have again man on a ship we might not even get a man we got a ship burning here let's try and up this we're gonna break the image probably but we might be closer to what we want see the image is broken but we have a man on the ship so you need to fine tune this like 7 to 14 ish is is usually a good number some models like to go lower now the clip skip here is a fairly Advanced setting and not something that you might want to play with if this is your first rodeo but think of the model as having layers so let's say that you want a person that will be your first layer now the second layer could be man or woman for example and let's say the third layer be say you have person Nam then it goes to for the third layer you can possibly be like some grandfather and it goes on and on and on like putting stuff into groups now this is extremely simplified it's not exactly like that but it's a good way of thinking and you might just want the person or a man but if you want to play around with that I suggest that you find a more detailed in-depth tutorial or Reddit guide for that this is not it the seed is a randomized generation so each time you will get a new image and a new noise if you want to recreate something you can reuse the seat so now we have this seed and that's locked to this image with well this seed and this prompt along with these Styles and all these settings will create this image and now you can recreate this it will give you the exact same image by using this you can have an image you can change the prompt or the settings to or fine tune it a little bit you're not gonna get like it's gonna be hard to create exactly this person with the different colored eyes I would recommend doing like in painting for that but it was good to know what the seed is so those are the basic settings of text to image if you're doing image image or what they call from image here you have a different set of setting mostly you have the same however you have one major difference which is the denoising strength and that's the most important part about image to image and I have a full tutorial on this you can check that in my on my channel but the denoising strengths how much the new image will retain from the all of them so let's let me show you let me put this to zero and generate now you will get the exact same image and if you put this to one you will get a completely new image so what this does is you can play with the denoising strength here and change parts of the image while still keeping some parts now we're fairly low here 0.3 so there will be changes to image but there are not really noticeable but you can see here like the light here is very different oh well not very but it's different from the one here this light is different this one and you know there are changes in the image if you would want to retain a lot of the image you have to go around the 0.5.6 but you still want the changes now you can see we're starting to get more change colorings in his face the markings are gone resembles the character but it's not exactly the same and then you can play with a value and now this is just a quick look at the image to image settings quickly looking over process here this is where you upscale mainly so once you have an image here you can resize it up to what eight times here but let's say you want to make it two times bigger you can change the upscaler and they differ slightly but uh honestly not a lot so in IR is great for architecture for example it adds a lot of fine details but it might be too much for some images for example faces I usually use LINE shows together with twin IR sometimes ESR Jam now we have an image that is upscale to time now it's not a great result because our image was just a test from uh 5 12 5 12. what you can do especially when your working image to image is you can change this to well 10 10 24 10 24. that way you can get a 512 image to go to 1024 retain some of the qualities of the older image so as you can see on this one the image is now fairly changed but we have more detail in the image as it's a double the resolution let's create a character here a woman having coffee in a coffee shop or Cafe same thing I guess and we're gonna add another style here from our Styles we're gonna add no you know what I'm feeling Ghibli for this we're gonna generate four of these as you can see we didn't get a live render preview here so let's see if we can find that in the setting in automatic 11 11 it was um here we have a live preview so this is the slider that we know it's set to -1 and this is we're gonna set this to one be able to see the full render here and if we generate again we should be able to see here there we go you can see all the images rendering and this is kind of interesting you get to see them together usually we'll start a setting maybe um here we are we show previews of all images generated in a batch as a grid so if you uncheck that just gonna see them one by one there you see if you go back you can see them well as we saw them previously as a grid so that's kind of cool if you're doing a lot of them now if you have a lot of models already installed you can go into the settings here if you check system paths here you can actually change to your automatic 1111 folder or wherever you have your models so this is where I am mine so I'm just going to apply the settings here and I can refresh and now I have a lot of models here we're going to use the colorful one now because that's fairly popular in my community and if you don't have any model you're going to be stuck with the 1.5 base model but I would recommend that you go to civitai download a model that you like deliberate for example then go inside your folder where you installed previously go inside models stable fusion and just drop them in here now let's do the same thing here woman in a coffee shop see here we have our woman in a coffee shop they're fairly low rest they're just 5 12 by 5 12. if you want to improve on these you could take for example this one here you have the little arrow here it says to move to that part we're gonna go to image which is our image to image and here we can change this to for example 10 24 10 24 we're going to change the denoising strength which is how much the image will change so there will be no change one will be full change we're gonna keep this at 0.6 for now now this is not a tutorial on the settings you can check my Ultimate Guide or more specifically like an in painting tutorial stuff like that if you want to know more about the settings and here we have our 10 24 10 24 version of this now it's not a perfect image in any way I wouldn't say but just from a quick generation I mean hey it's fairly good this is just to test out the tool anyway now if you find something like you can take this image for example you can either send this to in painting like in automatic 1111 we could pay interface here give that some detail I'm gonna change this to Woman face and we're going to change this to in paint area only masked which only to pick this again if you want to know more about the settings I have an in painting tutorial just check my channel it's going to be fairly easy to find if I remember also put the link up somewhere up top I guess you'll find it now we got some more details in the face here now it's still a very smooth out image but that's because of the styles that we used the digital painting kind of style if you want some more realistic you have to use some of the other styles now if you go into the settings again here you can find user interface down here if you go to the bottom here you have UI theme you click this little thing here you have a lot of different themes so let's see what we have here we have the default one I think you're all going to be fairly familiar with that one now we have to restart every time sadly to change the theme it's on the features list to be updated but here you can see this default theme so if you've been using automatic 11 it looks exactly like this either black or white let's check some of the other themes here's the glass one just gray and you have the little edge here as it looks little glasses a little Web 2.0 here's the windows one and this is very blue so if you're into blue this could be something for you check the steampunk you should like the steampunk cyberpunk kind of things this is alright with black and grams and yellow there on the important generate button now the cool part here isn't that there are beams but actually more than you can get anything that you want so here we go Radio soft now this is so I think this is more in line with what I would prefer you're gonna have to check this out yourself and see which one you prefer so let me know in the comments what you think do you enjoy that Vlad makes a quicker updates it has the extensions pre-built into the installer or do you enjoy the classic old automatic 1111 I for one we'll keep an eye on both of them and see which one gets ahead as of right now Vlad is making a lot of changes and updates to his fork and with automatic 1111 as of today which is the 16th of April hasn't updated anything in three weeks I guess we'll see what happens but the community is loving this so test it out let me know what you think in the comments below if you like this content like And subscribe if you wanna but if you don't that's fine too I'm not your boss do whatever you want as always have a good one see ya

Channel: Sebastian Kamph

Views: 39,849

Keywords:

Id: 8rMp-XXlSZc

Length: 19min 36sec (1176 seconds)

Published: Wed Apr 19 2023