CivitAI XL Model are HERE!!! + Awesome XL Hack!

Community trained XL models are here hello my friends how are you doing we're gonna check out an extension where you can load the refiner model at the same time as the base model then we're going to look at two different Community trained XL size models and then I'm also gonna show you a new hack on how to use the refiner model to do amazing photo bashing inside of image to image let's get started first let's have a quick look at the extensions so when you go into automatic 1111 there's an extension tab here you click here on the load from button and this will load all of the extensions that are available on top here you might have a search bar there you simply want to type refiner and then you see the refiner extension otherwise search in the list or use your browser search function to find the refiner extension then simply click on install this only takes a couple of seconds and of course as always afterwards you're going to go to install and then apply and to restart UI after the UI has restarted when you scroll down you see here a refiner section when you open that up you can enable the refiner and you can select from all of your models the sdxl refiner model then when you have selected this here you have a step slider but there is no denoise slider in here only a step Choice a classic Choice might be that up here for the main element where you do the base rendering you are setting 20 steps and then down here for the refiner step you're setting only five steps so that you have a total of 25 steps this also means with the new extension you don't have to switch between the two models you can simply click here on generate and everything will be rendered through the base model and then through the refiner and you will see the final image here on the right side so no more switching back and forwards between the different models of course alternatively you can use conf UI where you can also load both of the models here on the right side this is still faster in the render process and of course as you can see here with all of these different settings gives you a lot more flexibility and don't worry about all these notes and cables here you can download this I will link it below the video and here on the right side you see the text prompts for positive prompt and negative prompts so this is really easy to use on the right side here you have the Q prompt button and when you click that the image is rendered through the process when the image is rendered you can right click on it and then either open it in a new tab or save the image to your drive also because this is modular to make everything simpler you can simply take this text box over here and move it to the left side next to the image so you can simply have it in this View and then change the prompt and negative prompt here click on Queue and see the finished image here all also another thing I want to point out here is you do not have to copy the models into the comfy UI folder and the automatic 1111 folder you only need an extra model paths.yaml file right click and then edit I have a software called Notepad plus plus this is a free software if I open it up you can see I have here the locations of all my different sources checkpoints vaes loras and so on so you can simply copy in here the path where you have your automatic 11 11 models for example for the checkpoints and configs then save this and restart home vui next let's have a look at the two Community trained models one is dreamshaper which is a really good model and also very flexible in the different styles so when you look here at the examples you can see there is photorealism there is also this kind of digital art here there is Anime in here there's also logos you can do a lot of stuff with that now again I would highly advise for you to click on these images and then here on the right side you can see the prompt the negative prompted all the settings that have been used to create this image so you can create something similar and as you can see here I have used this prompt I'm getting also the dragon and it looks really beautiful another model I want to show you today is called hephaestos XL as you can see up here it can do both safer work and not safe for work images again you want to click here and look at the settings of what has been used in the prompts and so on so that you have a good start to know how to prompt for these models to get good results so now let's look at some slides I have prepared for you this is the heifer estos XL model base only rendered this is without the refiner I'm using Euler a as the sampler 20 steps 7 CFG scale you can see my prompt and negative property on the screen this already looks very good very nice and detailed now here's the same image but this is using the refiner at five steps with the refiner extension I have showed you before it gives more detail especially in the fur and makes it look more like a professional photography here we have the same image but now the refiner is at 10 steps although it's going to add more detail it also introduces some errors to the image especially here at the end of the baby lion cup you can see that the legs look kind of strange and the hip is kind of deformed so this is not as good as with five steps but another thing you can see here with the refiner is that in the base model the baby lion looks a lot more fluffy and cuddly so I actually like the base image more and with the refiner it of course looks very realistic very nice but a little bit less fluffy so I wish there was also a denoise slider so I could probably introduce a little more of that fluffiness here for comparison we have the sdxl vanilla model from stability AI with the same prompt and the same seed of course because it's different model still we're getting a different image here we have the same image with a refiner of five steps again we are adding some more details to the image looks very nice also and here we have a comparison between the sdxl plus refiner and the Hefe isos XL plus refiner model so you can decide on your own which of them you like better especially with this subject here the images look pretty similar but often I found that they have a Asus model creates a little bit of a nicer composition of course I also have created other images for you for example here we have a beautiful woman sitting in an art Decor room with gold and white so both of these images look very nice here we have another example where you can see a big difference XL Plus refiner versus Michonne 5.2 maternity 5.2 is very good with these Street scenes with a wet stream with these nice Reflections on the SD side it kind of looks like the refiner didn't work so there might be more experimentation needed this might also be down to the prompting here in comparison we have the dream shaper XL model plus refiner this is artistically more expressive it feels more fluent and alive so I like that scene more also I like the wet reflection on the street and here by the way we have a result that I have created with conf UI I have adapted the prompt a bit to have a wet straight to have people on the street so there is more going on and as you can see it has become a lot more similar to the mid-journey image although the details in the mid Journey image are crisper and also the street signs and neon signs look more real and here we have the dragon prompt from before again this has been created in comfy UI the dragon looks really beautiful here so this means that the community trained models can create really amazing images you just have to experiment with the prompting with your settings so to get the better results and now it's time for the hacker man showing you the forbidden fruit of AI let's go so my friends what I've done here is I'm using images from free pick as a source for photo bashing and then I have combined that with an alien planet with the moon and the background looks really cinematic but I want to render this with AI to turn it into a a consistent image so here's something that's a little bit strange but you have to do it because otherwise you're getting kind of an error message you want to load the sdxl refiner model go to text to image and render any kind of image you can use 800 by 800 or any lower resolution we don't need the image but for some reason you have to render this first and then send this to image to image now after this has done you can enter your prompt and negative prompt you can load the image you want to use you can set down here the resolution you want to create we use a high denoise value in this case I'm using 0.5 so that the refiner model can actually create something that is finer and more consistent where everything fits together in the image and this my friends is the result we're getting from that hack so this is an actual AI image that has finer details in there a little bit of a caveat here the Moon is egg shaped in the background I don't know why that's happening but it also has changed the landscape a little bit you can experiment with that here for comparison is again the original image and then again the image that I have created with the refiner in image to image render and that's it already for today my friends if you want to see more crazy hacks and tips on the news in image AI subscribe to my channel see you soon and thanks for watching bye oh you're still here so uh This is the End screen there's other stuff you can watch like this or that's really cool and yeah I hope I see you soon uh leave a like if you haven't yet and well um yeah

Channel: Olivio Sarikas

Views: 41,042

Keywords: oliviosarikas, olivio sarikas, olivio tutorials, stable diffusion, stable diffusion tutorial, ai art, stable diffusion ai, stable diffusion tips, sdxl, civitai xl, dreamshaper xl, HEPHAISTOS xl, comfyui xl, comfyui sdxl, a1111 sdxl, stable diffusion sdxl, xl model training, sdxl model training, stable diffusion xl, stable diffusion install, comfy ui tutorial, comfy

Id: V9lWhMKzT4g

Length: 10min 39sec (639 seconds)

Published: Mon Jul 31 2023