Create MULTIPLE Character Scenes – Midjourney Character Design

today I'm going to try to solve one of my Journey's Biggest Mysteries many of you have asked me whether it's possible to place multiple consistent characters within the same scene in mid-journey and I must confess I was not sure whether it was possible placing multiple characters in the same scene is one thing but making sure they remain consistent is a whole nother story naturally I had to find out whether it could be done and just so you know I'm not promising you anything but if you're curious to find out what fascinating things I learned then make sure you come along for the entire ride trust me you will not regret it so today I'm going to try to place two or more consistent characters in a scene I'm going to use nothing but mid journey and an image editor such as Photoshop or Affinity designer and I'm only going to use the image editor in order to prepare reference images I will not use it to create the final image so let's start off with attempt number one so for this first exercise I'm going to use our good old friends Carla Caruso and Peggy Palermo I will try to place Carla and in the middle of a New York City street and I also want them to be raising their arms in a fist as if they were protesting I'm going to start off by recovering the original reference image that I created for Carla Caruso in part 3 of this series imagine glamor Shot full body image beautiful woman bright red baited bun hair only red hair wearing a leather jacket in gray color wearing a white tank top under jacket wearing beige pants on legs in the style of a Marvel comic book an aspect ratio of two to three and if you have a real quick look this was one of the original character Concepts that we used in order to blend it into something entirely new then I'm going to create a similar type of full body character concept for Peggy Palermo as well imagine glamor Shot full body image on white background beautiful woman bob cut pink hair blue eyes black leather clothing black pants on legs and black boots in the style of a Marvel comic book aspect ratio of two to three so now that we have our source character Concepts let's create new images of Carla and Peggy in their protesting posts I'll start off by doing this for Carla imagine and then we start off with the reference image that we just created beautiful woman bright red hair braided bun and hairstyle only red hair Marvel comic book style and then holding a fist in the air raising her arm in a fist let's quickly check out what that looks like and these look pretty good so now let's do the exact same thing for Peggy imagine and we use Peggy's reference image beautiful woman bob cut pink hair black leather clothing in the style of a comic book holding a fist in the air raising her arm in a fist okay perfect these look really good so now that we have our character references let's create some background scenery I'm going to create a scenery that is in New York City and that is showing us a street imagine panoramic view of a Long Street in New York City dystopian theme Marvel comic book style and again an aspect ratio of three to two Okay excellent so these are just what I need so next I'm going to create three different image references inside my image editor in each one of them I'll use a different background image from this set of four and I'll also Place different versions of the characters on top of these in order to do this I had to remove all of the white background of the character references and then placed them on top of these background images and I'm hoping that these reference images will be dominant enough for my journey to understand exactly what I want so let's try blending all three of these image references into one image imagine followed by three image prompts and one URL for each one of the references yikes so these are just absolutely hideous okay so maybe I do need to add a text prompt after all let me add some more contacts and see whether that works imagine and then we start off with our three image prompts and Then followed by a text prompt that addresses our two different characters alright so let's have a look at our images okay so clearly something is not working here but what exactly went wrong well it seems that the image references are able to convey the correct composition but majority can't assign the correct hair to each character it also can't really assign the correct outfit and the faces are just plain ugly so this attempt clearly was a complete failure but maybe these characters are just too complex for this sort of thing I recently saw another YouTuber presenting his approach for creating consistent characters and he kept using very simple cute and small characters maybe it just works better with those however he was just creating character Concepts and he was not actually trying to place them within a particular scene either way let's at least give it a try moving on to attempt number two so let's start off by creating two different character Concepts first I'm going to create a cute little mouse wearing an ascot cap imagine character concept on white background multiple poses and expressions little white Mouse character wearing a brown Ascot cap children's book illustration in an aspect ratio of three to two so from this set of four I really like the bottom left one so let's go with that one and my second character is going to be a cute little bear wearing a red shirt imagine character concept on white background multiple poses and expressions little brown bear character wearing a red shirt children's book illustration so from these the bottom right one is absolutely perfect for this use case so let's keep that set next we need some background scenery for our characters as well so let's create an image of a children's room imagine ground level shot of a children's room children's book illustration style great so from these let's pick the bottom right one and then once again I bring everything back into an image editor I cut out my characters and then I place them on top of the background that we just generated and if it wasn't for these slightly different drawing style you could almost use this image as an actual illustration in a book now you might be tempted to Simply blend together the background image and the image reference that we just created and rest assured it will create an image that has a consistent style however it will also once again mess up our characters so instead let's try blending our reference image with a text prompt imagine and then we start with our reference image and Then followed by ground level shot of a children's room little brown bear character wearing a red shirt little white Mouse wearing a brown Ascot cap a bear and a mouse having a conversation children's book illustration style in an aspect ratio of three to two so this creates a set of images that do resemble the concept that we had in mind and the characters do actually look pretty good however they are not the characters that we actually wanted and their features are all over the place so what does this tell us well simpler characters are a bit easier to create placing two or more in the same scene is still very difficult mid Journey isn't able to assign the correct features to specific characters and you mostly get mashups of your ideas okay so as much as it pains me to say this I don't think that image references are going to get us very as far in this setup as soon as you want to use more than one specific character in a scene the lines start to blur and Concepts invariably start to bleed into each other much more than you actually want them to so this attempt was also a complete failure so now it's time to move on to multi-prompts and give those a try but before we do that I'd just like to note something as frustrating as all of this might seem this is exactly what prompting is all about we try to push the limits and in a collective effort we find out what actually works and what doesn't if you enjoy tinkering and experimenting with other people then come join us in the official tokenized AI Des Discord it's a great place to share insights help each other or just hang out anyway so let's move on here goes attempt number three multi-prompts have the major advantage of allowing us to reinforce certain Concepts from different points of view so that's why I'm actually quite hopeful that this might work however I also know that they do have a similar bleeding problem when it comes to using multiple characters of the same category this is particularly obvious when you try to place two different animals next to each other just have a look at this imagine a brown bear sitting next to a sloth okay so if you look at this image I think it's pretty obvious what the problem is here majority keeps creating weird fantasy creatures that are a blend of both animals I also have absolutely no clue what these strange bird-like creatures are doing in the image however things are considerably easier when you try to place two completely unrelated character types next to each other and by unrelated I mean animals and humans so let's start off with something relatively simple imagine Q brown bear speaking to a little boy so as you can see this concept is crystal clear to my journey but let's add some details I'm going to add some color to the boy's clothes imagine cute brown bear speaking to a little boy little boy wearing blue clothes okay so this works great but so far this is only a single prompt so let's add a second segment in order to reinforce the concept imagine cute brown bear speaking to a little boy little boy wearing blue clothes and that's the end of our first segment and then a conversation between a little boy in a blue outfit and a cute brown bear so as you can see whenever you add color to your prompt there is usually going to be some form of color bleed sometimes you can actually fix this by simply adding a negative prompt like this imagine followed by the entire prompt that you had before except that we're going to add a third segment which is a negative prompt for blue background now while this isn't perfect and it sometimes removes more color than you actually want to it's still quite useful to know how this works so let's change the location of the characters and place them on a farm imagine cute brown bear speaking to a little boy little boy wearing blue clothes Country Farm in the background and that's our first segment and then a conversation between a little boy in a blue outfit and a cute brown bear background scenery of a farm and that's our second segment so let me ask you a question do you notice the obvious problem in these images picking a farm was probably the dumbest thing I could do what do you usually find on a farm that's right a ton of different animals and that's why our bear looks like he's transformed into a pig-like or cow-like animal in the bottom two images once again medjuni fails completely at keeping two characters of the same category apart so clearly we need to move this to an entirely different location imagine cute brown bear speaking to a little boy little boy wearing blue clothes Garden in background that ends our first segment and then conversation between a little boy in a blue outfit and a cute brown bear background scenery of a garden alright so this is much better already now let's inject some children's book illustration style into the prompt by adding it only to the first segment imagine and then we repeat the entire prompt from before except that we've added children's book illustration to the first segment and here's what that will look like if you wanted an even stronger children's illustration style you would simply add it to the second segment as well so let's do that imagine we add children's book illustration at the very end of the second segment as well and here it is sometimes the increased abstraction of a children's book illustration can be quite helpful so I'm going to try something on a whim I really like the top left and the bottom left Styles so I'm going to try to apply these to a slightly different prompt let's start off by creating a new prompt without the image references imagine cute brown bear and a little boy sitting on a bench little boy wearing blue clothes Garden in background children's book situation that ends our first segment then little boy in a blue outfit and a cute brown bear sitting on a bench background scenery of a garden children's book illustration and that ends our second segment now you might really like the style of the images that you see here but please just ignore these for a moment the only thing that we're trying to do right now is to see whether a text prompt actually gives us the sort of concept that we were aiming for and clearly in this case it does so now let's try that very same prompt but add one of our image references at the very beginning of the prompt imagine and then we add our image reference as an image prompt followed by the same text prompt that we used before okay so this is actually something we can potentially work with let's try the exact same thing but with the other image reference as well imagine and then we use the exact same prompts except that we've replaced the first image prompt with the second reference image and we've also added a third segment which is a negative prompt to remove blue object this is supposed to prevent the bench from becoming blue in the image now I don't know about you but I think these images are absolutely awesome so I think it's actually fair to say that this attempt was at least a partial success it's crucial that you do not get too hung up about the minute details in these images ignore the Micro Focus on the macro if you plan on using this for a children's book let's be honest here the vast majority of the people out there are not going to notice these tiny differences as long as you get the basics right so what have we learned here avoid placing more than one animal in your scene one animal and a human is fine avoid color bleed by fine-tuning with negative prompts focus on getting the initial text prompt right then apply styles with image references and tweak your text prompt to change the composition and dynamic so what about three characters is that something we can do as well well let's give it a try this is attempt number four I'm going to skip the simple prompts and jump straight to a very complex prompt using three different segments imagine image of a garden a little boy and girl with blonde hair having a conversation with a white rabbit Garden in the background that ends our first segment then Garden scenery girl and boy speaking to a little white rabbit that ends our second segment and then a little white rabbit speaking to two children in a park and that ends our third segment okay so these actually look pretty good already we have an animal a female character and a male character I'm going to avoid adding color to The Prompt simply because I know that it's going to result in color bleed which I would like to avoid right now I'm also going to avoid adding any additional details to the characters because I know that it's going to lead to somewhat strange results for example if I were to introduce the concept of a dress then the boy is much more likely to have feminine traits or look like a girl all together instead I'm just going to add children's book illustration to the end of every segment of this prompt so we repeat the entirety of the previous prompt except that we've added children's book illustration on the first second and third segment okay so these actually look really nice I'm going to pick the bottom right image as an image reference what I'm going to do now is I'm going to try to apply the same logic that we used in our third attempt I'm going to start off by changing the prompt just a little bit imagine and then we repeat most of the prompt again except that what we've done is that we've changed the location by adding a pool and if we look at the images it gives us it's not entirely perfect but to be honest this is more than good enough for what we're trying to do so let's see what happens when we use this new prompt and add the image reference that we selected before imagine and that we add the image reference followed by our new prompt that contains the information with the pool and boom now that I believe qualifies as a success now I do realize that the clothes are changing slightly in these different images however you also know that you can re-roll these images these prompts as many times as you like until you do get the right combination and we know that this works but I actually want to go one step further I wanted to see whether I could keep the style and the composition but change the color of the character's clothes so I took the original image reference that we used as well as one of the new images and I applied strong colors to their clothes I then use both of these images as references and simply repeat the rest of the prompt imagine and then we start off with our two image references and then image of a garden pool little boy and girl with long blonde hair sitting next to a pool with a white rabbit garden with pool in the background children's book illustration Garden scenery with a pool girl and boy speaking to a little white rabbit sitting next to a pool children's book illustration a little white rabbit speaking to two children in a park with a pool children's book illustration and here's what I got now there's something very interesting going on here the boys clothes still aren't very consistent they keep changing in every single one of these images however I successfully managed to change the color of the girl's dress in not just one but all four of these images how awesome is that so looking back at all of our attempts what have we learned today the journey has difficulty assigning many specific details to different characters this includes facial features clothes and especially color color bleed is a recurring problem animals do not mix well with animals humans of the same gender also confuse maternity humans of different genders can work quite well in other words multiple characters require strict archetypical contrast image references can be helpful for style but only if your cortex prompt already works on a standalone basis simple characters are easier because abstraction tricks our brain into believing that they look similar this has been the most challenging problem that I've been faced with so far and I'm not going to lie it took me a full three days of pulling my hair out just to get to the point where I am now but guess what I now know how it's done and so do you this was part 4 of my character design series and once part 5 is released you'll find it right here in the corner of the screen if you missed the last episode go catch up on that content by watching part 3 right here and if you're new to the channel go check out the entire character design playlist right here if you enjoyed this video make sure to leave a comment keep on learning and take care

Channel: Tokenized AI by Christian Heidorn

Views: 78,705

Keywords: midjourney, midjourney v4, midjourney tutorial, midjourney prompts, midjourney scenes, midjourney storyboard, midjourney storytelling, midjourney camera prompts, midjourney consistent character, midjourney character design, midjourney character prompts, midjourney character creation, midjourney character art, midjourney weights, midjourney multi prompts, midjourney comic book, graphic novel midjourney, midjourney multiple characters, midjourney multiple subjects, midjourney v5

Id: znulbnRynOc

Length: 18min 23sec (1103 seconds)

Published: Sat Feb 11 2023