A Complete Guide To Loras In Stable Diffusion Automatic1111

Video Statistics and Information

Video
Captions Word Cloud
Reddit Comments
Captions
all righty guys hello and welcome back to another video on the channel today I'm going to be teaching you guys how you can do use luras in stable diffusion automatic one11 if you enjoyed today's video or find it helpful please consider leaving a like and subscribeing to channel as well if you're new okay so I've got my automatic one1 booted up here but firstly before we begin I'm just going to quickly explain the difference between a checkpoint and a Laura these base models checkpoints generally are more overarching kind of art styles they cover a more broader range of images and we use them to generate usually for example in a specific style like anime or realism or tune or like specific things whereas a Laura file does the same thing essentially but narrows it down to one very very specific thing that it's trying to make so say for example if you want to make I don't know images of Elon Musk for example so in today's video I'm going to be giving you guys an in-depth guide into basically using luras let's just jump straight into things first things first you're going to want to choose a Laura you're basically going going to come over to civid ai.com you want to click down here and go to models filter by most downloaded go over here to the right hand side click the filter button uh and uncheck everything except for Laura okay so I like this model this is a model uh based off the uh animated series Arcane that was on Netflix based off the League of Legends universe so we're going to choose this model here now a little tip if you ever need to figure out uh information about the model and how to generate the model uh you can always check the description and it'll tell you what is used in the thing sometimes but if it doesn't tell you here you can actually click on the images themselves and it'll give you the image generation data okay so I found a model I want to use with this Arcane Laura I'm going to try using this one here and see how it comes out so like always we're going to download the checkpoint first so go to your automatic 1 one1 folder and go to models stable the fusion and just save the file into there now do keep in mind yet again these files can be quite large this one's 4 GB so if you have slow internet it's going to take a while to download if you're fast into land like me it's not so yeah then we're going to want to download the Laura file so we're going to grab this we're going to hit download these files are usually a little bit smaller and then you want to go back out to your stable diffusion route directory you want to go to models and then you want to want to go to Laura and just save it into there and like always we're going to be using the ultra sharp model if you need to know where to download it just click the download button go to your R directory go to models and ESR again and but save it into this folder okay now that you have everything installed we're going to want to actually install low Pub if you didn't already follow my guide on how to install stable diffusion automatic 111 I'll leave a video link in the description below to check it out but basically you're going to want to install the lobe theme from this extensions you can just load it from file and type in lob and install it if it'll be there for you if you already have it um it'll be there for you if you don't already have it but basically you want to do is you want to come over here and this refresh button once you have this installed and you want to go to checkpoints and and it should be there so now we have the tune anime mix checkpoint just click on it and let it load up okay now that we have that installed we're going to want to apply our Laura we have the Arcane offset something to keep in mind is this information will be on every single Laura page but these usually trigger words for luras for example this one is this prompt tag here you need to generally use these in your models so a little tip what you can do if you have L installed you can click this little on your thing on this little settings window and it's going to bring this thing up so what you can do is you can click here press enter copy and paste that like promp in from here just by clicking on it and copy and paste it into here and now what's going to happen is uh sorry I didn't save it you if you click save it's going to add this every single time you click on the Lura so if I click on the low here it's just automatically going to add the prompt to the field box so let's just for example try it again it's always going to add it and it's always going to be there and if you ever want to format it just click this little magic one thing there so now we have that so now what we can do if we want to replicate one of these images or try to make one as close as we can as well we can take a character say for example we can take this character here and we can literally just grab this go like this and just copy and paste all of the prompts into the box and then we just get this little thing there I'm going to remove the uh Arcane offset cuz we already have that um and then I'm going to remove the Arcane Style just to make sure and then we're going to put in um all we're going to put in worst quality and low quality okay and then we're going to highlight each one we're going to hold control and press and then up twice while holding control and this is going to add the focus tags onto there I'm also going to use uh UL a 30 highres step uh 30 sampling steps so we're going to do a highres fix using the ultra sharp model we installed before and just going to do uh three times um and then we're going to do a highr STS of 30 as well and we're going to set the doo strength to 0.3 then we're just going to do 512 x 512 oops didn't me do that and batch count one batch size 1 and a CFG scale of seven Let's test and see how that comes out all righty as you guys can see this actually came out really nicely it looks like the character which is really really good it has all the details of jinx um it even captured some of the background elements of the underground Arcane city um all and all genuinely surprised how good this model actually is it's a pretty good one uh yet again links to all these will be in the description below if you want to check them out for yourself just to show you an example if I take out the uh The Prompt tag so if I take out the prompt tag and take out essentially the Laura I'm going to show you now what would happen if we didn't have the Laura in there now it might still get some details right because Jinx is a very very popular character that's the advantage of Laura it allows you to make really accurate images like this for characters that aren't as well known but let's see what this G this comes up with and as you can see it kind of got some of the details right for the character but it's not quite right uh there's it doesn't have like the shading and the line style of Arcane uh and it doesn't it just it just doesn't quite fit but it got like the the ponytail and the hair color right some of the tattoos but it doesn't actually like the face doesn't actually look like Jinx other image we got has the correct styling of that like shading that almost 3D 2D kind of look to it uh the face is more correct the hair is correct more clothing and tattoo everything's a bit more correct now I'm going to show you what happens uh let me just okay so now what I'm going to do I did I did this previous image this one here in the tune Style with the tune anime now I'm going to show you how you can turn them into other styles so if we use a model like epic realism which has a bit more of a realism style to it we can potentially take this character and turn it into a more realistic looking person and the way we're going to do this is we're going to add realisti to the end as as a prompt and I'm just going to hit the format button there now let's try that out and okay and here's the image in more realistic style and as you can see some of the facial details and the details in the image came out a bit weird but that's more just because this model is trained more for a tune art style so with some tweaking and some other potential things we might be able to get it to look right but as you guys can see this is how you do luras in stable Fusion automatic 101 to keep in mind these will work in any software that supports Loro so these will also work in voke AI it'll also work in comy UI if you wish to do it I'm going to show you guys a little trick so what want to do is you want to go go to load your extensions and you're going to want to type in this you want to type in civit AI into your extension field and the one we want specifically is C AI browser plus I want you guys to install this instinction once you've installed it just simply hit apply and restart UI all righty guys so say for example for whatever reason just a little tip that if you lost an image in your generation like you didn't get a chance to save it you can always just go into your stable diffusion rout directory and just go to the outputs folder text images and then find the D and then all of your image Generations will be in that folder now what we're going to do is I'm going to show you what this extension does so you're going to see a new tab up here called civit AI this will essentially rather than having to go to the website externally hit download download the Laura and all that kind of stuff this site is going to allow well this extension sorry is going to allow you to just download models directly into your stab Fusion install so what you can do is you can click here on this filter you can filter by model name username or a tag then you can go content type and we can type click Laura which we're going to look for another Laura basically that we're going to generate an image with um and then the rest of this stuff we can do just checking and out if you want nf up content you click that obviously um but I'm going to set the tile size to eight CU I want to make a few more tiles so we're going to set it to eight and then I'm going to set the tile count to let's say 50 just so I have a bunch more images I'm going to save that as default and already let's click off the filter and now what we can do is we just hit this little search button here it's going to search for a bunch of laurits just let it do its thing let it load and now as you can see it's given us a bunch of Laura options there's the pages over here so say for example if you want to go to the next page we can just click the next page and yeah so we can just cycle through a bunch of different Laura files and we can download them if we choose to so I'm going to go back to that first page cuz I actually saw one that I wanted to download I'm going to download the GAA Splash Laura now this is a Lura that I've used before in the past and quite frankly I really like it so there's some really really cool things that this extension essentially does so the first thing we're going to do is we're going to hit download the model so this is going to start downloading the model directly into our instore which is great so now if we go to Laura and we click refresh it's just there now you probably noticed a big advantage to this what it's done is it's actually gone from the C AI page and directly just downloaded this art for the cover art for the you know the model and this is really really good cuz it means you can kind of discern bit more quickly at a glance what each model is now as you can see before we have this information as you can see what it's also done is it saved all of the training data and a random prompt for the tags that we can do and why it's also good is you can view the Page information okay as you can see it didn't really work so this just happens sometimes in lauras as well so this is actually a great example of the show sometimes s two luras the combination of combining together they will not play nice together and apparently this is one of them in addition to that sometimes the model is not right either so that's a little bit unfortunate but there's not really a whole lot we can really do about that but what I will do is I will show you what the Gachet Splash art does look like so I'm just going to add the gacha splash art I'm going to make it isometric I'm going to add grassy field as the background prompt and then I'm just going to add the tag anime girl all righty so that's a little bit better that's kind of what the model can do the background image is a little bit weird but that would just take some generations and playing around to work but yeah guys that's basically a guide to Lura and hopefully you found my little tip of using the C AI extension to install Lura pretty helpful anyways guys have a good one see you later bye don't leave like And subscribe to the video if you're new
Info
Channel: Zifu's AI University
Views: 11,753
Rating: undefined out of 5
Keywords: A Complete Guide To Loras In Stable Diffusion Automatic1111, lora guide, ai image generation, ai image art, stable diffusion automatic1111 lora guide, stable diffusion automatic1111, sd automatic1111, a guide to loras, A Complete Guide To Loras In Stable Diffusion, ZifuAI, A Complete Guide To Loras
Id: zgJLJbzqYK0
Channel Id: undefined
Length: 11min 14sec (674 seconds)
Published: Sat Oct 28 2023
Related Videos
Note
Please note that this website is currently a work in progress! Lots of interesting data and statistics to come.