Comic Characters With Stable Diffusion SDXL

have you ever wanted to make comic books with stable diffusion what if you're a parent who has a whole bunch of kids and you don't have the time to sit there for hours on end drawing a background to make sure that it's perspectively accurate I've been an artist pretty much my entire life but there's one thing that I hated about being a comic book artist is the patience are you able to sit there for hours on end drawing the same thing over and over again it would drive me crazy well I'm going to show you how to use stable diffusion to tell the stories that you've always wanted to tell in comic book form now it's easy to get started with stable diffusion I've got an installation video that you can find in the description down below but aside from that we're going to be using the sdxl style selector and after detailer you can go into the extensions Tab and the available it should be there but if it isn't I'll leave the link for it in the description down below so you can copy that and paste it into the install from URL what I'm going to show you is how to create the images required to train your own lore model on a specific character but more on that later when it comes to trying to create a comic book and stable diffusion the one issue that we're always going to have is a consistent character and we've touched based on this in this video in the past but that was more for a realistic character for a comic book character we have the issue of the costume which is a bit of an issue because we want it to be consistent through the entire thing now the good thing is with a comic book and also with cartoons the Wardrobe tends to be very limited you kind of have Clark Kent either wearing a suit and tie or you have them in the Superman outfit you don't really have him being a fashionista going from look to look to look to look now obviously if you have Lois Lane you may have a couple different looks but for a specific let's say one comic book it's a good chance you're going to have her wearing the exact same thing through the entire comic book this is great for us because it means that if we train a character with one specific look let's say a superhero outfit or even something as simple as a pair of jeans and a t-shirt as in this case stable diffusion will have no choice but to actually draw the character with that look now in the past when we wanted to create a specific character we would have prompts that look like this foreign we're not doing that anymore because we're working with SD Excel the majority of these new models that are coming for SD Excel actually work a lot better with shorter prompts more precise prompts you just tell it exactly what you see in the picture for instance you would take something like this and bring it down to this it's much simpler so you don't have to constantly be creating paragraphs and paragraphs of description have you learned anything so far click like And subscribe that'd be great now the reason why we need sdxl star selector is so we can kind of direct it a little bit more towards what we want it's going to make things a little bit easier after detail is great because what it'll do is it'll impaint the faces on its own no matter how many characters you have in the scene so what I want to do now is I want to create a female character with blonde hair gonna be wearing leggings and a tank top and she's going to be running through New York as you can see I've selected a name from the random name generator that's exactly what we did in the last one when we're trying to get consistent faces you're going to find the link to that in the description as for the other settings here's a list of the other settings and what I like to use you are not restricted to these settings for instance the sample method you can use whichever one you like there's actually quite a few that I like to use the next thing we'll do is go to after detailer activate that leave everything else as it is today we can generate the image [Music] all right I love it it's perfect so you see we kept it nice and simple and it gave us the exact image that I was looking for one thing to keep in mind if you tell stable diffusion to put your character in a section of let's say New York most of the images that it's been trained on has a lot of pedestrians in New York so what's going to happen you're going to have a lot of pedestrians in the background if you've told stable diffusion to put a specific look to that character that you want in front and center it may spill over into those characters as well and you'll see that certain characters in the background will be dressed exactly like your character you may have to impaint those out later that's fine that's easy we can do that okay so now we've got the image that we want but we're not settled down on a seed which is fine because what we want is we want to generate a whole bunch of these one thing to keep in mind when it comes to buildings they're not the greatest things to actually generate with stable diffusion there's quite a few lines and there's a lot of perspective to keep in mind it tends to struggle like the windows tend to get a little bit warpy sometimes it happens sometimes it doesn't because we haven't locked off the scene it's going to randomly generate a whole bunch of images so we'll click generate and see what happens [Music] foreign as you can see with all of these the face looks exactly the same on all of them that's the benefit of using a specific character name as opposed to just saying one woman if you say one woman one man it's just gonna randomize it everything's going to look completely different in every single picture we've got all the images we need for this data set now the thing is in some of them it doesn't look like the same character for instance the pants might look a little bit different this is where it helps if you have a little bit of artistic skill but it's not that necessary you can use in painting and kind of you know work your way through it sometimes in paint might not help you so knowing how to use Photoshop and how to draw a little bit is definitely going to be a key here I'm going to throw this into a photo editor and alter it just a bit I don't have to go into crazy detail with this I'm not drawing this thing from scratch all I'll need to do is draw the line and then change it up a little bit make sure that I kind of go with the lights and shadows of it it doesn't have to look perfect it's good to actually use the Clone stamp when you're using Photoshop that way you can kind of get the same style with what you're changing around I'm going to save that out now I'll bring it back into stable diffusions in paint and I'll select mask only I'll draw over the section that I want and click generate depending on your denoise number it'll keep the exact same image I would keep this around 0.5 if you're trying to integrate something a little bit better sometimes it's easier just to use Photoshop or some sort of photo editor like krita or photo P even then actually going back and disable the fusion we're actually building the photos required to train a model so they need to be the same in every single image if you have something that looks a little bit different it's going to start getting confused it'll give you results you don't really want in the final model if you're going to be creating superhero suit with an emblem on the chest I would suggest that you actually put those in in Photoshop and then that way you can use the warp tool to kind of move it around and put it on the chest and try and fix the lighting to match what it is that you've got in the image that way you know that from image to image it's going to look exactly the same then we can send that into in painting put the noise into 0.5 and just paint over the logo you'll see that it actually embeds it perfectly into the image you didn't have to worry about going through like you know 50 different images just to fine tune two logos that match because if you try and do it straight out of stable diffusion they're not going to come out good you're going to get logos that look insanely different from picture to picture it's actually just easier to create an image in Photoshop it's going to match every single time because you've done it with Photoshop perfect that way when we train our model the actual logo is going to look exactly the same as if you had a Superman logo because you know the Superman logo looks exactly the same every single time you try and generate a disabled Fusion or in this case you're going to have your own logo pre-trained into the model and that way it'll be perfect every single time in the next video I'm going to show you how to take this data set that we've created today and turn it into a Laura file using koi SS and possibly some online options if you don't have a GPU that can support it if you guys have any other questions let me know in the comments down below if you liked the video click like subscribe and also check out this other one foreign

Channel: Sebastian Torres

Views: 17,396

Keywords: stable diffusion, promting, stable diffusion prompt guide, stable diffusion tutorial, ai art, ai, beautifull ai art, talk to stable diffusion, characters, comics with stable diffusion, consistent faces, consistent characters with stable diffusion, stable diffusion comics, how to make comics with stable diffusion, After detailer extension, How to make comics, Lora datasets, Comic book artist, Consistent characters, Ai comic books

Id: eCDcTjjWW8I

Length: 7min 57sec (477 seconds)

Published: Fri Sep 15 2023