The Tricks to Consistent Characters in Midjourney Version 5

so you're looking for the secret to getting consistent characters across different scenes in mid-journey I've seen a lot of bad advice out there and I just want to tell you there is no secret there is no feature built in a mid-journey that allows you to do that however there are some tricks that you can use and I'll show you them today let's start with generating our character and the first lesson is to prompt a human-like subject because photorealistic humans all look the same whereas a cartoon can be drawn in many different styles also keep your design simple keep your prompt simple the more complicated Your Design the Lessman journey is going to be able to carry over from scene to scene full body portrait of an Instagram model detective style flowing silver hair by using the word detective I kind of chose the fashion of the character and then I think mentioning their hair color and style is worth adding to your prompt so you're going to have a couple options to choose from number two is the only one that has a little too much going on that would be hard to replicate in other pictures like this camera strap and the colors it's not great but the others in number one we have a dark silver Blazer and a white shirt underneath we have the trench coat and number three with a tie and glasses and in number four again we have a silver trench coat and a white blouse I think so we upscale them but here's the first big trick I want you to hit variations on the pictures you like we are going to be using these upscales as an image prompt but I think we need to be using more than one picture and we're gonna get that through the variations also a quick note I've seen some really bad advice out there don't let anyone tell you that hitting favorite is gonna somehow teach mid-journey what you want reacting to your images does absolutely nothing it's a placebo effect it's a myth it's an urban legend I've seen a lot of big YouTubers recommend this it doesn't work it's it's it's not true never happens it's totally made up pure fiction now what are we looking for in the variations I'll make it really simple you want something that looks different enough in this example do any of these stand out I don't think so I'd say they're all 99 the same we're looking for something that's like 94 the same just a little bit different is going to make a big difference with these variations as well I don't see enough of a difference and you know what while we're here on this topic check out this green V3 right here what that means is that I turned on the remix feature and I tried it out because when you hear about the remix feature you might think oh I can take a picture I like and remix it into something else you can change their pose you could add them into a different scene and it's just it doesn't work how you might think it would I remixed that photo by adding in triumphant pose hands raised in air and look how bad these pictures are it does it does not work maybe I just don't know the way to use remix but as far as I can tell nobody has a down pat so here you go here are variations from one of the other pick pictures and what do you notice in number three is just enough of a difference she's looking in a different direction and that subtle change of her face I think matters a lot it gives me a journey more to work with so we'll have a picture like this and a picture like this and what we're going to do is click on it so it expands then we're gonna right click on it and hit copy image address now in our prompt box we can hit Ctrl V or command V or you could right click and hit paste but we're going to paste that link into the box now it's going to be this big Long messy link don't worry once you hit enter and generate it it's going to shorten it up and make it a lot easier to read after you copy and paste those links you're going to go back to your original generation and you're going to copy that prompt we're going to be using that as the basis of our new prompt and then here's where we get to experiment and add different words add different scenes mix things up a bit so there's what the links look like when they're shortened and then I stuck with full body portrait of an Instagram model but this is where I mixed it it up I went with relaxed lying down with hand resting on chin then the rest of the old prompt detective style flowing silver hair now it didn't do exactly what I wanted but it's still that character in a different scene kind of a different look I'd say number three is not bad number four as well is kind of a cool picture you can basically get a full photo shoot doing it this way but it's not precise it's not like control net you can't manually pose these characters right now at least but what other tricks can we use well there's a couple things with regards to where we place the words in our prompt Instagram model detective style blowing silver hair sitting on a park bench so we changed it up and we put it at the end of our prompt and what do you think are these okay I'd agree that's the same character same clothes same hair pretty similar face but you know they're not the best pictures so we could add some weights to our proms so we have Sitting on a park bench and then I place two colons and then added a weight of 1.5 so when your image prompt something you have your links and then you have your prompt when you add double colons and a weight you're adding a weight to your prompt versus your pictures so having her sitting on a park bench was 1.5 times more important than the reference images and I think these look pretty good but adding any little bit of a weight probably helps here it is with just a weight of 0.5 and I think these look pretty good I like number one a lot number three is not that bad either number four the anatomy it's not quite there her arm's a little wonky but there is one more trick you can try not just adding a weight you can also add some chaos to the prompt and what this does is it changes the variety of the grid you're given without any chaos each of the pictures are going to look pretty similar so if you get a bad generation they're all gonna kind of suck but with chaos one of the four might be pretty good and that's what we're aiming for sitting on a park bench 1.5 weight s 400 C4 now you might like number three that she's looking away there's a higher chance you really find what you like but I'd be careful at C40 you might not find anything you like it's going to stray from your prompts quite a bit like number one is just that's a scary picture and none of them are quite her anymore so be careful with the chaos value but let me introduce you to the first problem you're gonna run into and this one is hard to solve I changed up the prompt and I put buying groceries I wanted her in a grocery store but because our two reference images are her facing the camera that is what you get here it's not natural it's not a candid pose of her in a grocery store and it honestly doesn't matter what you do to the prompt I even put behind shot of an Instagram model buying groceries still nothing she's facing the camera you can also use the slider method which is when you take a part of your prompt that's being ignored and you slide it to the end while adding a weight there behind shot of an Instagram model that's fine but then we have behind shot again with a weight of 0.8 and it kind of works in number three just kind of you'd have to experiment with this one a lot it's not a golden solution here it is at 1.5 and it's a little bit better number one and number three aren't bad but number two she's staring right at the camera just like the reference images another thing you could try is lowering the stylized value which means mid-journey will follow your prompt more closely however it's not perfect I mean number one's not bad and you notice the camera in all these pictures if you negative prompted no camera that might help here is the negative prompt no camera lens it gets rid of the lens and like number two is pretty good but maybe her hairstyle is not quite right so honestly you could hit variations on it or maybe number four is good enough for you as you can see here there's no easy solution for getting your character in different scenes just a lot of little tricks you can try so you've noticed how our reference images are kind of holding us back here's one thing you could try for that you could try using the phrase character sheet in your prompt as well as the words multiple expressions and poses and you'll get something like this this works much better for animated characters or you know cartoons where you can then take these pictures put them in a Photoshop separate them and place them into different scenes that might work really well especially for cartoon characters but for here I don't know what do you think it's not bad I mean you definitely have a character that you could place in different scenes like in number one and number three she's dressed down a little when a number four she's got her tie-ons loosening a bit maybe she's getting a little frustrated you can definitely see some story elements here again you could add some chaos and see what mid journey is really capable of number three there's some usable poses there it's really not that but I don't recommend C40 you're probably not going to get what you want there's a chance but it's not good when you don't know what you want maybe that's when you can use a higher chaos value but anything specific keep it low and then I tried getting this character investigating a crime scene I went through all the tricks I've tried so far adding a bit of weight to what I wanted in the prompts didn't turn out that good I tried changing up the camera angle wide angle shot again not that great I tried bumping up the weight and you know maybe these are a bit better but still she's just looking at the camera this one I think worked out really well here is where I divided the prompt into three different parts the first part wide angle shot of an Instagram model inspecting a crime scene colon colon which is going to separate this from the rest then we have detective style flowing silver hair colon colon and then we slide over what we want inspecting a crime scene 0.8 look at number two and look at number three that's almost perfect that is basically perfect that is our character in different scenes what should we call her detective Mallory here's detective Mallory in a couple different scenes but here's the big trick I want you to try put simply you're going to generate your background separately and then you're going to include that in your image prompts CSI crime Scene dramatic lighting so we get something like this I liked number three a lot because you can see where a person would fit into that picture so we're gonna take this we're going to click on it we're going to right click and hit copy image address and then we're gonna put that into our prompt I'll like number one a lot number two is not bad but she's still facing the camera you see how there's no real way to do this perfectly and then here's a quick example of using version 5.1 full body portrait of an Instagram model I recommend keeping the design simple so I went with minimalist clothing short purple hair and I think number four is going to be the easiest to replicate so we're gonna hit upscale but we're also going to hit variations so we can get more than one picture but this time it didn't really work out I don't think any of these are similar enough but different to really help in the generation so we'll just stick to one picture for now and I tried every trick in the book it's really hard to get this character in different scenes trying to get them driving a white car doesn't work even with the slider method really no success using the multiple expression trick actually works out pretty good if you were to take number three into Photoshop divide it up and then use the back of her head in an image prompt that might work out really well trying to get them to walk down a busy City street not really impressed so what do we do we generate create the background separately like here we're gonna upscale number two and then boom here's our character in The Busy City street but do you see how none of those are actually heard none of them are the character we wanted that's because generating consistent characters in different scenes isn't really possible it's not built in a mid-journey but I've shown you some tricks you could try and for the record using the blend feature which is forward slash blend and then just choosing the two pictures which being our background and our character worked out a little bit better that's her in the streets now and instead of using multiple expressions in your prompt you could try something like being specific and saying side profile and then getting one of these pictures or maybe slide it and do slide profile 1.8 and I think number four and number one are really good there you have it it's not an exact science but those are the tricks I would try to get consistent characters in different scenes I don't want to oversell it to you I really don't recommend doing this because it's not an exact process it's a lot of trial and error but maybe that's what you're looking for and this would be fun for you to try I hope you're doing well take care and I'll see you next time peace

Channel: Future Tech Pilot

Views: 16,404

Keywords: midjourney characters, midjourney consistent character, midjourney consistent character design, midjourney consistent character v5, midjourney consistency, midjourney prompt tips, midjourney tricks, midjourney tutorial, midjourney character tutorial, midjourney beginner tutorial

Id: ICh9ose7tGs

Length: 12min 5sec (725 seconds)

Published: Mon May 08 2023