Face Swap GRAPHIC NOVEL Characters from Midjourney

today I'm going to embark on a journey through Uncharted Territory that's right I'm going to show you how to use stable diffusion to face swap characters in your mid-journey images now before you click away I'm not going to do this with just any majority images I'll be doing it with characters that are in a classic graphic novel illustration style and that my friends is not as straightforward as you might think sure there are certain illustration Styles where inside face swap works perfectly fine and I even demonstrated this in one of my most recent videos however when it comes to illustration that leans far more towards line art then basic face swapping just doesn't do the trick now swapping the face into your images is pretty straightforward however maintaining the style of your images that's a whole different story most of the time your swap faces will lean far too much towards photorealism and they won't really look like their illustrations at all anyway by now I think the majority of you have gotten used to the fact that the vast majority of my videos only cover my journey that was never my initial plan it just naturally evolved this way because it was the image generator that I simply enjoyed the most so as I record my very first video today about stable diffusion I am absolutely prepared to be ridiculed and mocked in public that's fine go ahead just bring it on and I will admit I am nowhere near the level of expertise of some of my YouTube buddies who specialize on stable effusion I do use it for certain things every now and then but to be very honest it's just a little bit too nerdy for me however I do like to think that I'm pretty good at connecting the dots and finding Creative Solutions to problems so some of you might remember this video right here that I recorded a few months ago where I demonstrated how you could create your very own comic book with the help of a few AI tools I use chat to BT to spin up a very simple story and I use mid-journey to create the imagery that we needed for the panels of the graphic novel I then packaged the whole thing rather nicely so that it would appeal to a wider audience on YouTube and of course the inevitable happened the keyboard Warriors showed up hidden behind their Anonymous usernames and their sense of superiority they had nothing better to do than to pinpoint every single tiny little mistake error or inaccuracy but I guess that just what they do the primary complaint was that the characters did not look consistent throughout all of the scenes and yes that's a limitation that I actually pointed out at the beginning of the video but hey why pay attention if your primary objective is to complain anyway today I'll show you exactly how you can fix these problems in your very own projects and best of all it won't cost you anything so first things first for any of this to actually work you're going to need a few things to be set up on your system you're going to need a working implementation of stable diffusion on your computer or in some other Cloud environment in this particular video I'm going to be using one of the most popular ones which is automatic 1111. so if you don't already have that installed make sure you check out this video by my buddy Sebastian and he'll walk you through the entire process in a soothing and calm voice now once you've installed automatic 1111 you'll also need to install the rube extension and once again I recommend you check out Sebastian's video on how to do this as well this video already covers part of the process however we do need to take some additional steps in order to make this work and trust me those are the secret sauce to getting this to work with graphic novels now let me show you what we'll be working on let's assume that we created a similar adventure story to the one that I did a few months ago Treasures Beyond gold and the graphic novel that we're creating uses an illustration style that is very similar to these images here's the prompt for the main character Leo imagine blend of comic book art and line art in full natural colors portrait of an attractive and muscular man in his early 30s with short cropped brown hair and a stubble beard tropical rainforest in the background and then I add a hint of beige just to get certain color tone Oh and by the way I'm using version 5 here because this is where I managed to get this very particular look this produces the following type of shots and I decided to go with the top right one this headshot of Leo will serve as the source image for the face swapping process next we need some Target scenes into which we can swap Leo's face this is more or less the same prompt except that I've changed the composition by adding a full body shot and I've also added details regarding his clothing so from these four Images I picked one that I liked the best I then made some additional Minor Adjustments to The Prompt just in order to change the scene and from these four shots I picked the bottom right one as my second target image I then decided to re-roll the very same prompt one more time just in order to get another variation of this scene and from these four shots I decided to go with the top left one this time okay great now let's move on to our Maya character here's the prompt for her headshot imagine blend of comic book art and line art in full natural colors portrait of an attractive Asian woman in her late 20s with an athletic build and shoulder length dark hair tropical rainforest in the background this produced The Following image grid of four and I really liked the top left one so I decided to go with that and just like I did with Leo I created additional shots of Maya with similar camera angles and scenes so whatever you're working on you'll need to create an original headshot of your characters and you'll use that as a source image and then you create all the different scenes that you need for your storyline but remember this process will only swap out the face so really just the facial features that means that you still need to make sure that all of the other details are roughly in line with what your character looks like so for example the hair color or the hair style and you'll also have to pay attention to your character's body as well as the outfit all right so now that I've created all of my image assets let me walk you through the entire process once you've prepared everything for stable diffusion and loaded up automatic 1111 this is what your interface will look like hey sorry to interrupt you right smack in the middle but I just wanted to share something with you YouTube is an absolutely amazing way to discover and learn new skills just like the video that you're watching right here however as you discover these videos and tutorials you're never quite sure when they were recorded whether they're still relevant and how they fit into the bigger picture of learning to use and control midtermee so if you're looking to get a comprehensive training in the foundations of prompting in my journey and I highly recommend that you check out my video course at masters of mcjourney.com which gives you a comprehensive education in learning how to control my journey and how to prompt with intent now I'm not going to go into all of the details and intricacies of how stable diffusion Works especially not automatic 1111 I'll leave that to The Experts such as Sebastian or Olivio and I'll focus on what you need to learn in this process so first of all you make sure that you've selected whichever model you want to use up here in the top left corner and then you go to the image to image tab once you reach this page you then click on the in paint tab over here and then you select your target image and I've prepared all of these images but so let me select my first Target image which is this one I select it so since we're in painting here I now need to select the area of the image that I want to change so I need to apply a mask in order to do that I'm going to adjust my brush size a little bit here and then I will simply select all of the characters image sorry not image face like this and it kind of depends on what the face looks like in this case it's very important that I cover the entire face because Leo's Source image has like a full beard so now that I've selected the part of the image that I want to change I can scroll down and here I need to select in mask mode in Peyton mast in this case I think it was automatically selected then I move over to the in paint area and here I select only masked now from the sampling method I'm going to select DPM plus plus to M Karas to be honest I'm not too familiar with all of the differences of the different sampling sampling methods however I think this is just what I saw Sebastian do and this worked pretty well then next for the sampling steps I'm just going to leave it at 20. this will mostly determine how long the process takes then for the resize 2 we can leave that as 5 at 5 12 512 for the batch count I'm just going to reduce it to 1 and the what batch size I'm also going to leave at 1 and next for the CFG scale I'm going to set this to seven and for the denoising strings I'm going to set that to zero because we're just swapping in the face now next is very important the part about group The Rope extension down here you basically open up this toggle then you select the image that you want to swap in in this case this is my Leo Source image here you see it and don't forget to enable the extension over here and then you've pretty much done everything that you need to do we can now simply scroll back up we go over here to the orange button generate and I click it and all we need to do now is wait for stable distribution to do its work in the background oh and by the way the roof extension uses the exact same model that is used by inside face swap the original onnx file which is kind of like a model used to be available on the official GitHub of inside face swap but they subsequently removed it after they realized just how popular it was and since they've started to monetize it now you I guess you kind of know why now the good thing for us is that the file had already sort of spread across the internet and it comes automatically bundled with the roof extension anyway so this generates an image that looks very similar to what you would get with inside face swap and you can clearly see that the swap face has facial features that are very very realistic and they don't really look like an illustration or a graphic novel of any sorts but guess what it's surprisingly easy to fix this and I'll show you how it's done all right so let's go back to automatic 1111 and see how we can fix this so first of all we're back at the image to image Tab and the in pane Tab and this time I'm going to now remove my initial Target image and I'm going to load a new one and the new one is essentially just my first result so you can see this is the one that we and painted with a face swap now once again we select our mask tool or in this case I already have the right size and this time we're going to draw over the face again and we're going to select the exact same area this time and let me just get this right this should do the trick or less and then we scroll down we don't need to change a lot I tend to switch this to 10 sampling steps don't ask me why this is just kind of what I've experimented with it's quicker than using 20 and it produces pretty much good results then next what we need to do is we need to change the guidance scale and the denoising strength now this part is really up to you you need to kind of experiment with what works with your particular style in my particular case I had pretty good results using anything between 7 and 9. and a denoising strength of 0.4 anything above that or below that led to certain changes in what the face would then look like and they didn't really feel like the Fidelity to the facial features were good enough but this is kind of more or less The Sweet Spot that I found for my images you'll maybe find different ones for yours now a very important part here is we need to deactivate sort of disable the Rope extension because we don't want to swap in the face anymore right so you just hit the enable tick box here so that's disabled or unchecked and then once we've done all that we can scroll back up and we do one little tiny detail here graphic novel or graphic novel illustration it doesn't really matter and to be honest I'm actually not entirely sure whether this is absolutely necessary it's just what I kind of started to do and it had an impact so next I'm going to hit the generate button and we'll wait for this to generate great and now let's have a look at how this has changed our image so what this additional step does it takes all of the primary features of our swapped face and then simply applies a graphic novel Style on top of it you can now see that Leo's face doesn't just have the facial features of the original headshot they also have the right style and you can do this with all of the other images as well let me walk you through a quick slideshow of all of the images remember this is Leo's headshot we then take this target image which looks nothing like him and we swap in the source face the only problem is that it's too realistic but with the additional step we can turn it into the right Style here's another example with all three shots side by side pretty neat isn't it so let me show you how this looks on Maya here's the original headshot for Maya which we'll use as a source image and here's the first comparison you can clearly see how the facial features have changed and the final image looks like it was taken Straight From the Same Book here's another one by the way you can repeat the last step over and over again until you get the right look for the face so you're not stuck with just one of these image Generations this is really important because it doesn't always work perfectly straight away but since this is stable diffusion it won't cost you a dime here's one last example that I want to share with you it's important to realize that face swapping only works as well as the target image that you choose the closer the overall anatomy and the style of your target image is to the source image the better the space swapping process will be now I do realize that many of you use my journey on an iPad or some other type of setup that can't run stable diffusion but if you still want to try this out there are a number of service providers out there that basically allow you to use stable diffusion in a cloud service since I haven't actually tried any of those out I can't really make a recommendation however I'm pretty sure that if you ask Sebastian or Olivio they'll have tons of recommendations on what type of service to use I hope you found this useful and maybe this will help you get one step closer to finishing that project of yours and remember there's a whole bunch of free stuff and courses in the description so check that out and I'll see you next week and take care [Music] thank you [Music]

Channel: Tokenized AI by Christian Heidorn

Views: 10,105

Keywords: midjourney, midjourney v5, midjourney version 5, midjourney tutorial, midjourney prompts, midjourney workflow, midjourney learn, midjourney 5.2, midjourney face swap, midjourney insightface swap, stable diffusion face swap, automatic1111 face swap, midjourney comic book, midjourney graphic novel, midjourney consistent character

Id: G-c7BlEyuP8

Length: 15min 28sec (928 seconds)

Published: Sat Aug 12 2023