Want Consistent Characters in Midjourney? Do THIS Instead…

Video Statistics and Information

Video
Captions Word Cloud
Reddit Comments
Captions
it's kind of been mid-journey's ultimate Challenge to truly generate the same character in any scene you want if you tried to do this before you know that most strategies fall apart the moment you want to change your character's expression if you want to put something in their hand or add more people into the scene my method that I'm sharing in this video can do all those things and it takes place entirely in mid-journey you don't need Photoshop you don't need to train Some Dream Booth model and to start doing this yourself all you'll need to do is watch this video to the end hi my name is globetry and I make videos helping you navigate the landscape of AI experiences today I'll start by showing you how you can put this character Elara into absolutely any scene in a photorealistic style regardless of the pose the lighting the composition of the scene or what she's doing Elara looks stunning every time and more importantly she looks the same every time once you understand the principles I use to add her into a scene I'll be able to show you how you can do the exact same thing with any character in any style based on a prompt you can write yourself if you end up liking the video or liking my vibes well I hope you decide to click the like button because that's what it's for and it helps me out a ton so I'd really appreciate it anyway here goes if you want to generate another version of Alara in a new scene or doing something we've never had her do before the workflow is actually going to be very easy first we need a prompt for the idea the scene we want to put our character in I don't like wasting your time so I have one prepared right here in this case it is a photo of a medieval lady in a deep emerald green dress kneeling and lighting a bonfire in a vast Courtyard with gray cobblestones towering stone walls and a hint of a castle all illuminated by the Warm Glow of rising flames Wireless generates I want to point out that this prompt only has a simple description of our character throughout this whole method I'm showing you in this video you will never have to write some big complicated prompt that describes every physical trait your character has just simple description I promise you this is easy enough you will actually be able to do it you might have to watch the video again but you would do that anyway just to support me because the video is just that good oh look at that it's generated obviously this looks nothing like our character this is Alara this is this obviously I didn't even mention the correct hair color but that's okay you just gotta trust the process so now step two in our workflow we will use my new favorite feature very region this lets us erase all the parts of the lady that we want to change and because this doesn't look much like Elara yet at all we're going to be changing most of her let's just draw around a Lara right here I'm going to leave a bit of her hand a bit of this so kind of her dress might end up looking somewhat the same but I'm mostly removing everything we've got here I want to show the anatomy of her leg and go like this essentially these are the things that mid Journey needs to figure out the lighting and the pose and since hopefully I wrote a good enough prompt to start with these parts here look enough like Elara that we shouldn't have to do this more than once now as always we're gonna have have to write a prompt this prompt will need to include everything we need to make sure that a Lara looks correct watch how easy this is in here I will type dash dash e l a r a dash dash Alara this of course is the prefer option I set up to make sure this goes as smoothly and as easily as possible now I'll add a little bit just a touch of detail Elara is let me paste this in kneeling and lighting a bonfire illuminated by the Warm Glow of the rising flames and now we click generate you'll see here my dash dash Elara got replaced by more than a prompt but actually a combination of an image prompt and a text prompt so what I have here is five reference images of the character you'll notice they each have slightly different lighting angles and poses these images here help ensure that mid Journey knows exactly what I want her to look like and because mid-journey prescribes equal weight to text prompts this text here a photo of a medieval Martian girl with red hair a green dress and an off-white blouse this part is to make sure the image prompt does not conflict with the text prompt and it gives us the highest chance of generating at least one variation that is exactly what I want so yeah doesn't she look great when I check to see if she really is the same and not just close I've been paying attention to specifically her pointed chin her bluish eyes the size and roundness of her nose the fact that she has high cheekbones and while the color of her hair changes the touch when the lighting is different the texture of her hair should stay exactly the same looking at these right now I'd say that both of these Generations are actually a really great generation of her I'll pick one to upscale and yeah that's ilara bluish eyes round nose pointed chin high cheekbones and gorgeous flowing red hair with just a little touch of frizz I've had this work on dozens of images of my character I showed you these during the intro but they are absolutely stunning and it only takes those two steps generate an image with the full creativity of mid Journey's best creations and then use very region to replace the features of that character with the features of my character this method is generic and now I can put Elara into any scene I want even in totally different styles or genres now I could give you these five images or even this whole prefer option which I am doing it's in the description but that will only let you have an easy time generating my character I'm sure the reason you're watching this video is because you want to start generating your character just as easily and I'm sure you've noticed that if you want to start doing this yourself you're a little trapped you're in a bit of a catch-22 because you need to have a fleshed out image prompt to make this worked but if you want to generate a bunch of mid-journey images of the same exact character basically what I'm showing you how to do you need to already have a bunch of mid-journey images of the same exact character seems kind of impossible so what you really need to know is how I generated these without having them already before I give you step-by-step instructions what I need to say is the this will be an iterative process my character becomes better and more consistent the more I use her because I can replace and update the images in my slash prefer option with ones that represent my character better but the real answer is a smaller more iterative version of what we just did let's start broad Strokes how do you get a consistent costume design and basic features well I tend to start by generating a character design sheet like this with a series of poses I find I get the most interesting designs when I generate them in a cartoon watercolor style so here is my prompt for that let me paste it in and you'll see it is a watercolor painted cartoon style character design sheet of many poses for blank consistent character from many angles and this time we'll do a purple alien just to show you how generic this concept is and that you can do it for whatever character idea you have so we'll paste in an alien creature with purple skin tiny ears and fluffy black hair we'll see how this looks I added a little quote there from when I pasted it and generate the image when it's finished generating we'll see that it'll give you a few options and now all we have to do is pick one I know it doesn't have tiny ears but this one in the top right is adorable so I'll go upscale number two and then I will take this character sheet and make a bunch of screenshots of the poses I like and paste them right back into the mid Journey bot now I'll take all of these image URLs and paste them into a slash prefer option for my prompt we'll call him purple these images will be our starting point there'll be a rough guide for mid-journey so it understands what purple looks like and create something relatively similar each time let me add a little bit of a text prompt as well and now you can see I have a full slash prefer option with all of this information in it now you can see that if I do slash imagine dash dash purple like this we get this character that looks basically the same as the screenshot in this nice watercolor style at a full resolution even though the concept of purple alien is relatively vague and could come up with a lot of different things this alien that we're generating now looks similar across all four of these images not only can we just generate purple like that but we can also do slash imagine purple on the moon slash imagine purple sitting on a park bench or slash imagine purple scared and hiding behind a mailbox all these images should come up looking somewhat like purple and somewhat in some of these different scenes each time he remains stylized and he has pretty cool features the moon and the park bench as well as the mailbox might not be super consistent but each time he is our alien purple with his version of tiny ears his version of purple skin and his version of black fluffy hair but we want to generate these in a new style don't we so let's go ahead and try to make him exist in a photographic style copy this right here slash prefer option set a photo of an alien with purple skin tiny ears and black fluffy hair will take these same images and put them in front and still even though these images are in a cartoony style the style of the image will come from the text prompt and hopefully we get something that looks at least almost photographic when we generate slash imagine purple can also generate these same images again purple on the moon purple sitting on a park bench purple scared and hiding behind a mailbox let's time travel wait for these to generate now with these images it's becoming more obvious that we have work to do to properly Define our character in the eyes of mid-journey sometimes it's too cartoony sometimes it's too strange sometimes it's not obvious what he looks like at all so now that we have some photorealistic versions of purple we need to drill down and figure out how to take our consistency to the next level the way to do that is to work feature by feature getting everything you need to line up I know this is already a long video so we'll just do one feature and then talk about the steps you'll need to create your character to be as well defined as Elara is so for purple I'd say his eyes need the most work these ones look good but some of these look totally cartoonish so how are we gonna do this what I would do is take the two images where the eyes are pretty much how I like them these ones here so now what I'm going to do is choose the eyes that I like the absolute most and take a screenshot of just those so my screenshot will be just of purple's eyes right like this rather than taking a screenshot of the whole character now we'll paste that screenshot into the mid Journey bot get our URL and now click on very region for a version of purple where we don't like the eyes very much so we'll click very region right here and now I can write a quick prompt describing the eyes paste in the URL and erase them from the image Up Above This Way mid-journey can sample the eyes I provided and update them accordingly you can see here the eyes aren't perfect but they're no longer the cartoony Madness that they used to be and they look a lot better now this version of purple obviously isn't great and I probably won't be using him in future proms but maybe this one stands a better chance we'll click very region again paste that same prompt we just wrote and erase purpose eyes right here and you can see this one here is also a huge Improvement now we have three photos of purple where all the eyes are the same so we can update our prefer option with new URLs grab this guy this guy and this guy since they all have eyes we actually like ah why not we'll do the crazy guy too and notice this time we have all four images in here but we're not describing the eyes themselves instead we're simply trusting the image prompt to give mid-journey all the info it needs we'll only add eye specific information if we're having real problems replicating the eyes we like but I have a feeling we won't need to do that anymore let's generate a new image of purple how about Slash imagine purple riding a hoverboard sci-fi city backdrop generate him now remember these are all relatively photorealistic so we'll probably create a better photorealistic style and they'll all have very reasonable eyes with this purple Hue interesting Reflections because those are in the image prompts we provided while this generates I want to point out we can repeat the same process for his mouth his nose his ears or anything else where we're losing consistency and we can get better more refined and more interesting images as as we go now purple already is starting to look a lot better look at these highly reflective glossy eyes on what is becoming more and more a consistent character we can do this over and over again with new traits and new features until purple becomes better and better I have to tell you this will take some iteration and you'll need to really figure out how you want your character to look but in time by showing your character some love with a lot of very regions and trial and error you'll eventually get to a place where you have five or six high quality images that show off your character extremely well in a diverse set of poses and angles and at that point the sky is the limit you'll have a two-step process for putting your character into any scene you can generate if you end up creating a character you love I hope you share it in my Discord and maybe I'll highlight it in an upcoming video my Discord is also a great resource if you're having problems getting your slash prefer option working or trouble with any mid-journey generation at all I'm sure there's someone in there who can help you get unstuck we've got a lot of really talented AI artists I do want to thank you all for watching I know this was a pretty Advanced guide in it I kind of assumed you already have a good understanding of how to use very region and how to write a prompt that generates photorealistic images with different poses and lighting but if this stuff is new to you or you just want to keep learning I have another video for both of those things you can watch this one for a quick guide to Mid journey in painting while I create a monstrosity or this one where I give you my strategy for creating a great impactful photorealistic character whichever you choose I'm sure you'll have a blast see you there
Info
Channel: Glibatree
Views: 51,148
Rating: undefined out of 5
Keywords: GPT-3, AI Demo, AI Story Telling, AI Art, Midjourney, ChatGPT
Id: lLo4BB64vUk
Channel Id: undefined
Length: 16min 41sec (1001 seconds)
Published: Wed Oct 04 2023
Related Videos
Note
Please note that this website is currently a work in progress! Lots of interesting data and statistics to come.