Best Practice Workflow for Automatic 1111 – Stable Diffusion

hello my AI art generating friends in this video I show you what in my personal opinion is the best workflow for stable diffusion in automatic 1111 and give you useful tips and tricks and insights today for model we are going to use ref animated so it's a semi-realistic model it can do very beautiful renderings you should try it out if you haven't it's very cool what you can do with it I set it to clip skip 2 and I always say I don't show it again but I actually show you guys because you're asking me in the comments anyways so for clip skip for the slider to appear you go to settings from here you go down you click user interface and then you click here into Quick Settings list click enter here and you type in clip stop at last layers that's it how that's how you get it after that you need to restart your UI I've already prepared a prompt for you it's a scene with a female astronaut and a space station and an exploding space station in the background I will share the prompts as usual on my patreon page so let's get over the settings real quick we use Euler a Euler a is very good for prompt engineering then when you're in the face where you're experimenting around use Euler a because it's fast then the width and the height is 768 is the maximum what most models can handle otherwise you get deformations and you will see horrible things so we don't want to go higher and 432 is because I want to 16 to 9 resolution for my screen so if I want to render a wallpaper or something so I crank up the batch size to 8 because we want to select some nice and images from here okay and now we render let's see yeah this one is kind of nice this one also yeah they all have quite a potential this is cool I'm stuck between this and the last one so I would say we go with this one I like the explosion the best I think so this has this cool Sparkle explosion in all directions users let's work with this a lot of people now use Cyrus fix we don't do that because Iris 6 can change a lot in the composition as you know from my other tutorial video and in this workflow we want to change the sampler and keep the composition we use send to image to image because we want our finished image we want to be based on this we want we don't want a completely different image now we change the sampler we change it to DPM plus plus 2m arrows so also I resize the whole image by two now you will be surprised what it does to the details like the face and for the denoising strength you want to set a value between 0.4 and 0.7 0.7 if you want a lot of changes 0.4 if you want almost no changes we want to introduce some changes we want to let us be surprised I set the batch count to three I like to have a bit of choice let's see let's see what we got yes this is quite nice so yeah I like the explosion this is also good yeah we have three nice images but we have to choose one and now for our next step we go to intent we send it over as this video has been helpful to you subscribe to me mean a lot to me what it does for me is it gives me an indicator that I'm building an audience and that I'm creating helpful content for people and now what do we want to do we have this kind of leg situation here the leg down here is coming from below and this lag is actually missing it's it's disappearing that's not good we have to fix this for girl so she needs another leg now we Over Paint her her face and go to settings only mask this is important and for scale now what I do is I go by 0.5 so I basically have the resolution to what it was before so the trick is since we down scale it gets down sampled so we don't lose that much detail we scale it up later so it gives a very nice and crisp image for denoising strange let's set to 4.5 we actually have a good phase we don't want to change it that much just a little so as you can see even though we scaled it down it still looks very good and I'm quite satisfied with face so in this we get the most of the eye I think ah there's no huge difference between two and three let's take this here so if we send it over to intent again and now he deleted the mask but I can tell you who didn't delete the mask click here up here again that's important because now it says start drawing if it's not says start drawing then there is still the mask I have made this error a lot of times I cannot count actually or many times I've made this error so now let's try to give her the correct lag so let's fix this error because she has to feed obviously and Two Boots so let's give stable effusion some room here and for this I would actually set up some denoising strange to 0.6 so because we want to change some more of the image yeah we have these three images I don't know what that is said doesn't look healthy does look broken in multiple ways so this is okay this is good this is yeah good ish um no let's I think let's go with this one so here is a little robot her thingy or maybe no it's not a glove um yeah it's it's a robot of thingy I have decided that now for the final step we send it to extras and here from the upscaler menu I choose the r s rugen 4X anime 6B I use this often for images I render with ref animated um because it's a semi-realistic look so use this and you're gonna be good there may be other options there are huge discussions on the internet which upscaler looks the best with Which models so uh feel free to discuss along it's not my cup of tea I'm fine with this yeah look at all the details so it looks so beautiful the explosion everything the details are perfect um I really like what we got here I have a lot of new videos lined up I have one tutorial about in painting hands the seven common mistakes people make with automatic 11 11 and many more videos for you so stay tuned

Channel: AIKnowledge2Go

Views: 183,744

Keywords: AI News, Artificial Intelligence, AI Tools, AIKnowledge2Go, AINews, chatgpt, News, Automatic1111, Tutorial, Learn, RevAnimated, Civitai, Workflow, Image2Image, Inpainting, Upscaling, Upscaler, Inpaint, AIartTutorial, LearnAI, HowTo, SDXL, SD XL, Stable Diffusion, artificial intelligence, stable diffusion lora, stable diffusion, automatic1111, stable diffusion img2img, aiknowledge2go, stable diffusion prompt guide, stable diffusion tutorial, prompt engineering, inpaint anything, automatic 1111

Id: RiN-zrUlneQ

Length: 7min 59sec (479 seconds)

Published: Mon Jun 26 2023