New Method for Midjourney Character Consistency (Not the cref method)

Summary

In this video, the creator shares a method for achieving character consistency using the AI tool 'Mid Journey'. They discuss their journey since version 3 and their quest for consistency, which they feel they've nearly mastered. The tutorial focuses on using the 'ni6' model for its cartoon and illustration aesthetics, avoiding the common 'CF method'. The creator outlines their process, which includes using various poses of a character in a single image, style reference images, and character reference images to guide the AI. They detail steps for refining the image, repainting, and reprompting to achieve the desired character consistency, offering tips for creating a library of character images for future use. The video concludes with the creator's reflections on their method's effectiveness compared to the 'CF method' and encourages viewers to like and subscribe for more content.

Takeaways

• 🖌️ The presenter shares a method for achieving character consistency using the AI tool Mid Journey.
• 🔍 They've been using Mid Journey since version 3 and have been experimenting with character consistency.
• 🎨 The ni6 model is preferred for its aesthetics, which are better suited for the presenter's cartoon and illustration projects.
• 📸 The method involves using various poses of a character on the same image to achieve consistency within a single image.
• 📝 The importance of a well-crafted prompt is emphasized, including character sheets, multiple angles, aesthetic directions, and style references.
• 🖼️ The presenter uses three types of images in their prompts: silhouette hints, style references, and character references.
• 🔄 A repainting process is described for refining images, where the AI model is guided to focus on specific aspects of the character.
• 🎭 The process involves removing prompts related to multiple poses and focusing on a single action for the character in subsequent repaints.
• 🗂️ Creating a folder in Mid Journey to store and organize images of the character as a growing library is recommended.
• 🔍 The presenter suggests using the reframe tool to change the aspect ratio and perspective of the character in the scene.
• 🚫 A caution is given about the limitations of the CF method when trying to generate multiple characters, which tend to look the same.

Q & A

What is the main topic of the video?

The main topic of the video is achieving character consistency using the AI tool Mid Journey, specifically with the ni6 model.

Why does the speaker prefer using the ni6 model over V6 for their projects?

The speaker prefers the ni6 model over V6 for their projects because they favor the aesthetics of the ni6 model, which is more suitable for their illustrations and cartoons.

What is the 'CF method' mentioned in the script?

The 'CF method' is a technique where multiple poses of the same character are used alongside a descriptive prompt to encourage Mid Journey to produce similar characters or poses.

Why does the speaker emphasize the importance of having various poses of a character on the same image?

The speaker emphasizes having various poses of a character on the same image because Mid Journey can achieve consistency within a single image but struggles with consistency across multiple images.

What role does the aspect ratio play in the process described?

The aspect ratio plays a role in determining the number of poses that can be effectively displayed side by side, with the speaker finding the 4x3 ratio ideal for the proportions of their character.

Why is style consistency important in the method described?

Style consistency is important because it helps to ensure that the generated images look cohesive and belong to the same character, which is crucial for projects like books, comics, or videos where visual consistency is key.

How does the speaker suggest improving character consistency in Mid Journey's outputs?

The speaker suggests improving character consistency by using a combination of character references, style references, and repainting techniques to guide Mid Journey towards generating more consistent character images.

What is the significance of repainting in the method described?

Repainting is significant as it allows the user to manually adjust and refine the generated images, focusing Mid Journey's attention on specific areas or details to achieve the desired consistency and accuracy.

How does the speaker handle the process of adding new actions or backgrounds to the character?

The speaker handles the addition of new actions or backgrounds by repainting the desired area and then reprompting Mid Journey with specific instructions, allowing for the creation of new poses or settings while maintaining character consistency.

What is the speaker's advice for organizing and storing generated character images?

The speaker advises creating folders on Mid Journey for each character, storing images that are considered consistent and useful for future reference, thus building a library for further development of the character.

What unexpected feature does the speaker discover about the 'CF' command in Mid Journey?

The speaker discovers that when using the 'CF' command to generate multiple characters, Mid Journey forces all characters to look the same, which can be leveraged for maintaining consistency in character design.

Outlines

00:00

🎨 Achieving Character Consistency with AI

The speaker introduces a method for achieving character consistency using the AI tool mid Journey. They have been using it since version 3 and are continuously seeking ways to improve character consistency. They mention using the mid Journey website, which is familiar to Discord users, and prefer the ni6 model for its aesthetic appeal. The speaker clarifies that this tutorial will not cover the CF method, which involves using multiple poses of a character to generate similar images. Instead, they focus on a technique that uses various poses of a character in a single image to achieve consistency. The speaker emphasizes the importance of the images used in the prompt and how they influence the AI's output.

05:00

🖌️ Refining the Character with Repaints

The speaker discusses the process of refining a character generated by mid Journey through the use of repaints. They advise patience and attention to detail, especially in maintaining consistency across poses. The speaker shares their approach to cleaning up messy elements and ensuring that character traits are consistent. They also explain how to adjust the prompt to focus on a single pose and action, removing elements that suggest multiple characters. The goal is to achieve a character that is consistent in style, proportion, and detail, which can be used across different projects.

10:02

🚀 Expanding Character Possibilities

The speaker explores how to expand the character's poses and settings by repainting and rerunning the AI generation process. They demonstrate how to change the background and action of the character while maintaining consistency. The speaker also shares a tip for creating a folder in mid Journey to store and organize images of the character, building a library for future use. They mention an experimental method of generating a background without repainting around the character, which they plan to explore further.

15:04

📸 Framing and Reframing Characters

The speaker talks about using the reframe tool to change the aspect ratio and perspective of the character in the scene. They discuss how editing the prompt during reframing can influence the generated image. The speaker also shares an accidental discovery regarding the CF command, which forces multiple characters to look the same when used in a prompt. They demonstrate how this can be used to create multiple poses of the same character in the same setting.

20:05

🌟 Conclusion and Future Exploration

The speaker concludes by reflecting on their method for achieving character consistency with mid Journey. They compare their approach to the CF method and express their preference for their own technique. The speaker acknowledges the support from the Reddit community and expresses their enjoyment in creating the tutorial. They encourage viewers to like and subscribe for more content and mention their plans to continue exploring and refining their method.

Keywords

💡Mid Journey

Mid Journey refers to a specific version or iteration of the AI tool being used in the video to create illustrations and cartoons. It is central to the video's theme as the speaker discusses their method for achieving character consistency using this AI tool. The video mentions using Mid Journey since V3, indicating a progression and improvement over time.

💡Character Consistency

Character consistency is the goal of maintaining uniformity in the appearance and attributes of a character across different illustrations or frames. In the video, the speaker is chasing the 'unicorn of character consistency,' which implies the challenge and desirability of achieving this uniformity. The entire video is dedicated to sharing a method to accomplish this using Mid Journey.

💡Aesthetics

Aesthetics in this context refers to the visual style or appearance of the illustrations produced by the AI. The speaker prefers the 'Aesthetics of the niji model' for their projects, indicating a personal preference for how the characters should look. Aesthetics is a key consideration in the method being shared, as it influences the style of the generated characters.

💡Style Reference Image

A style reference image is used to guide the AI in producing artwork that matches a certain visual style. The video mentions using a style reference image to 'bind them together in some language,' suggesting that it helps the AI understand the desired artistic direction. This is crucial for maintaining a consistent style across different character illustrations.

💡Character Reference

A character reference image provides specific details about a character's appearance, such as features, clothing, and accessories. In the script, the speaker uses character references to ensure that the AI-generated characters have 'some general traits' in common, which is essential for maintaining character consistency.

💡Prompt

In the context of AI-generated art, a prompt is a set of instructions or descriptors that guide the AI in creating an image. The video discusses crafting a prompt that includes 'character sheets, multiple angles,' and other descriptors to achieve the desired outcome. The prompt is a critical component of the method for generating consistent characters.

💡Repaint

Repaint in the video refers to the process of manually editing or refining an AI-generated image to correct details or to guide the AI in the next generation. The speaker mentions 'repaint and the first thing, you'll need to do is clean it up,' indicating that repainting is a step for improving character consistency and details in the artwork.

💡Aspect Ratio (HR)

The aspect ratio, referred to as HR in the script, is the proportional relationship between the width and the height of an image. The speaker mentions using 'the HR of 4x3' to influence the AI's layout of character poses within the image. The aspect ratio is an important factor in the composition of the final artwork.

💡CF Method

The CF method is a technique mentioned in the video where multiple poses of the same character are used alongside a descriptive prompt to guide the AI. The speaker notes that they did not use this method in their approach, but it is mentioned to provide context on different techniques for achieving character consistency with AI tools.

💡Reframe

Reframe in the video is a technique used to change the perspective or composition of an image. The speaker talks about using the reframe tool to 'change the aspect ratio and positioning' of the image, which allows for different viewpoints or scenes to be generated while maintaining character consistency.

Highlights

Introduction to the method for character consistency using mid Journey.

Preference for the ni6 model over V6 for its aesthetic appeal in illustrations and cartoons.

Explanation of the limitations of the CF method for achieving character consistency.

The importance of using various poses of a character on the same image for consistency.

How to use a style reference image to bind character poses together.

The significance of the image aspect ratio (4x3) in producing ideal character poses.

The process of refining character consistency through multiple repaints and reprompting.

Techniques for cleaning up images to maintain character consistency.

The strategy of selecting one pose at a time to create new character actions.

How to modify the prompt to focus on a single character action without multiple poses.

Using the lasso tool to select the character for repainting without affecting the background.

The impact of prompt adjustments on the generation of character backgrounds.

Building a library of character poses for future reference and development.

The discovery of a 'flaw' in the CF command that can be exploited for character consistency.

Comparing the results of the CF method with the presented method for character consistency.

Practical advice on how to manage and organize character images for future use.

Exploration of additional tools like reframe and zoom for enhancing character images.

Reflection on the process and the creator's personal experience with character consistency in AI.

Transcripts

hello everyone welcome to the veil of

virtue where we explore our creativity

using AI tools available to us right now

today I'll be sharing with you my method

for character consistency using only mid

Journey I've been using mid Journey ever

since V3 and ever since I saw that first

blob on my screen I was amazed and I've

been chasing the unicorn of character

consistency ever since and this week I

feel that I got as close as I ever did

so stay tuned I will be using the mid

Journey website since most Discord users

can and know how to use the website and

the other way around it's not

necessarily true I will also be using

the ni6 model because as you can see

here in my gallery most of my projects

are illustrations and cartoons I know

you can achieve cartoon and illustration

using V6 but I just prefer the

Aesthetics of the niji model this will

not be a video about the CF method and

if you watching this video you probably

know what I mean by that it's a method

where you use a lot of poses of the same

character alongside with a prom that's

descriptive of that character and hope

that mid Journey can produce similar

characters or similar poses of that

character I also like to use a style

reference image to try to kind of bind

them together in some language uh and I

did use and abuse this method and I did

got some cool results but as you can see

here they are not consistent enough to

call it the same character or at least

to call it call it the same character

side by side on the same project as you

can see the proportions aren't right the

art style changes even details like her

nose is different every time and you can

call it the same character at some in

some projects but Mo for the most part

you can't like have a book or a comic

book or a video and say it's the same

character the first thing you'll need

for this method is various poses of your

character

on the same image because as you can see

mid Journey can't give you consistency

across images of the same character but

it can do so on one image and will take

a advantage of that you can use your own

method to achieve this if you got one if

you have played with character

consistency and if you use mid Journey

for a long time you probably have your

own set of tools and ways of getting

things done but if you want to copy my

prompt you'll see it's fairly easy at

least it works for this kind of

character character The Prompt the words

in the prompt at least are character

sheets multiple angles and I can change

that varying on character I may

experiment then I just give it some

aesthetic directions and finally the

style but for me the most important part

of this prompt are the images I'm using

and the way I'm using them and I'm going

to break that down for you I'm using

three images here the first image I'm

using is this one and I'm not using it

as a style reference or a character

reference just just a link just an image

and for me what this does is it hints

the model on the silhouette of the

overall image I want it helps to tell

the model that I want the same character

in different poses side by side for this

effect I also use the HR of 4x3 you can

change this but for me this works the

best because I feel the model always

produces or tends to produce three poses

which are ideal for the proportions I

have on my character because space will

be a factor and because space will be a

factor if I have more character

references they will take up less space

and less details but if I have less for

example two I might lose on reference

because I want to keep referencing at

least two poses of the

character style is very important to me

and that's why I use a reference style

and here I'm using just one image for

the example but I can have as many as 6

7 8 or even 10 because what I'm trying

to do here is tell the model that the

images it should produce should be

inside that style and the more you have

the more consistent images you have a

same across the same style as reference

the less of a surprise will be the style

that appears on your generation and

that's what you want control over the

the style you're

using the same can be said for the

character at this point you should have

an idea of the overall image you want

for your character some traits that it

it must have so using a character

reference is very important if you want

to have a saying on what your character

will look like obviously the character

reference will not be the same as the

character you produce otherwise this

method the CF method would be perfect

and you wouldn't need this video but as

you can see some general traits are the

same the coloring she has glasses she

has a backpack some of the features you

already must know about your character

are there and please make sure that when

you get to this stage you do like the

character you end up with because this

is the start of a process that can take

a long time depending on the project

you're working on and you should enjoy

doing this with the character you have

if you're not happy just please keep

rrolling a good um hint would be to try

to very subtle or very strong where it

will take your image if you're kind of

satisfied with satisfied with the traits

of your character but not the overall or

you want to see some variations just

keep hitting that button and see what

comes up because most likely there will

come up some more characters in three

poses in the same sheet with some

characteristics that are already there

and you have a library to choose

from and that's the basis of the prompt

I'm using but keep in mind that this

might be different for the art style or

character type you're

using when you're satisfied with a an

image that has different poses of the

same character that you are willing to

use just it repaint and the first thing

you'll need to do is clean it up a bit

be patient and pay attention to the

details that aren't that good for

example here I selected this hand that

looks very messy and this shoe where

it's missing the gray stripe you want to

keep an eye for things that either aren

consistent withit uh across poses across

characters here you want to keep it

consistent so if you see see something

even if it's not looking bad but it's

not consistent across the the character

for example here I could repaint this

handle or do the same here because here

the backpack doesn't have any and here

it

does and for example I was talking about

this shoe selecting this gray stripe and

things that look bad like this hand and

you can at this stage you can just leave

the prompt the same and just submit you

might have to do to do this a few times

to get something that you're happy with

and here you can see the end is looking

better the shoe as the gray thing and I

could go into more detail but this is

more than fine for the project I'm

working in not just for the video the

the project overall this is as close as

a character consistency I have ever

worked around using mid Journey so I'm

pretty I'm pretty happy with

it so the next step would be to hit that

repaint again and now you will select

one of the poses a full section of one

of the poses here I did the middle one

but you can take any of the poses you

want and after doing that you should

reprompt it first of all as you can see

I took the um image link not the image

reference or this the character

reference but that just image link I add

to hint the model on the silhouette we

don't need that anymore and I repr

prompted everything that was concerning

the multiple poses angles character

sheet everything that was hinting that I

wanted more than one character I took it

out and changed it for an action here

eating an apple I was able to keep

everything else the same like this

direction in style the sub the SRE the

CF but keep in mind that some actions

are much harder to get out of the model

than others I have no idea why maybe it

has something to do with the database

but if you're having troubles try on

your reruns try to keep taking out

things so that you can emphas emphasize

what you really want the model to do for

example you can end up with something

just like this and even so even when you

do that sometimes it's really hard but

in this case it worked so let's hit

submit and see what we

got and there you have it you have your

character on on a new pose doing

something you asked for consistent in

style in proportion in details in color

and you can put it side by side with

other versions of that character and

call it the same character to me this is

amazing but now your character is is

standing next to two poses of itself and

that's not what you want so what we'll

do next it's hit repaint

again inside repainting what you'll want

to do now is Select everything but your

character with the lasso tool selected

try going around your character as close

as possible without going over it if you

do go over it we will you will force the

mid Journey model to reimagine parts of

the character and now it won't have

those two other characters to reference

itself because they will be

repainted still you got to get as close

as possible because if you leave huge

chunks of this white background M

Journey will be forced to integrate

those on the new image

and it will it won't make as good as a

job creating a background as you can see

here I've changed the prompt where it

said eating an apple where where it said

the action the character should be

playing now it has a description of a

background here I simply put at a Bach I

was able to keep everything else the

same and I also took the CF out because

the only character you want in the image

it's already there and now just hit

submit and here is the result and I got

to say I'm amazed with this because you

have your character in a post that

wasn't there in at the beginning you

chose that your character would be would

be eating an apple you chose that your

character would be at a Bach and this is

not only character consistency but as

close as as to some kind of control as I

feel I ever did on M

Journey again keep in mind that probably

it won't get you won't get something

like this at first try you have to rerun

it a few times here you can see I

repainted the hand now she's it looks

much better than it did before you can

see here and I'm sure I repainted a more

a lot more of small things that I didn't

like just clear your image a little

clean your image a little just give it a

nice wash before you go to the next

stage which is adding things and to add

something you just keep following the

same method as you probably already

figured it out you would repaint here I

chose this area and I changed the prompt

to pirate ship I was able to keep the

the CF but I probably probably had some

trouble getting the pirate ship to

appear because as you can see here all

that flat 2D design cartoon I took that

direction out that aesthetic direction

to emphasize that I needed a pirate ship

and I usually do this when I'm rerunning

and I'm not getting what I want from the

model then you just hit

submit and there you have it a pirate

ship or at least a ship it's not

necessarily a pirate ship but the cool

thing is that with this method if you

are patient enough you can go in detail

and repaint here and rerun it a few

times and maybe had a flag well you can

add in theory you can add or remove

anything you want from your image if you

are persistent enough to produce new

images just repeat the process you can

use any three pose image of your

character that you want you don't need

to need to use the initial one because

every time you repaint a region and the

new character and new pose of your

character appear appears here even if

it's not what you ask for for example if

you ask petting a dog and the girl just

appears here just doing nothing as long

as the pose of the character is

consistent with the other two and you

like it you can use that as a new

reference for a new pose also REM

remember that you can select any of

these three poses and in my experience

if you select for example a pose where

the character has a back turned the the

new image that will produce will also be

kind of with a back turned which can be

a good thing because it gives you a bit

more control over what you can get of

your image and control is always a good

thing in AI

Generations after you choose the pose

and you select you read reprompt that

area for the new action you get your

character in doing that new action you

must you probably will have to rerun it

a few times as I said before but when

you're happy where with the new pose you

can now use that image and repaint

everything around your character

everything but your character please do

a much better job than I did here

remember what I told you about uh mid

Journey having to incorporate this white

background as you can see here she has

mid Journey finds clever ways to

incorporate this background that you

live but the less it has to do the

better and now with this selection you

can prompt that repainting for your

background and that's how you get an

image and now you can clean it up add

things but that's the basis of

it and after you repeat the process a

few times you'll end up with a set of

images of your character in different

places doing different things and we can

go over some

things that went well and others that

didn't do so well for example here I

didn't notice at the time that she was

missing her glasses that would have been

an easy fix when she was next to other

references of herself but now it's

really hard to produce the same kind of

glasses of course I can incorporate that

on the storytelling and say she lost her

glasses or something like this but keep

an eye for those inconsistencies because

it can drift make you drift away from

your character here I was trying to have

the girl looking down at a frog but I

when I repainted I used an area that was

too big for what I wanted and mid mid

Journey used it because that's what it

do it does it plays a lot with the space

it has so I should have just repainted

this area probably and

not that whole area and that ended up

giving me a frog this big this is on me

not the model and you can uh do better

than this

and I'm sure there's still a lot of

inconsistencies but you can be much more

thorough during the process to avoid

them and even at this stage this is

something I would feel comfortable using

on a project like putting this images

side by side to tell a story and call it

the same character and here is another

example of using that method to achieve

character consistency it's fairly

consistent not not just in character but

in style this style is very different

from the other one I using I was using

before and as you can see it's still

very consistent you do have this

character which is this pose which is a

bit more complex but I think you can

make like a closeup and get away with it

and this is the simpler one so the most

distant two are this one from this one

but apart from that they're all fairly

consistent just a piece of advice I wish

I would have followed which which is

every time you get to a an image of your

character or poses of your character

that you do like and you think you'll

you will be using later to further

develop the character either through

making new poses or new backgrounds just

create a folder here on Mid Journey just

go to create folder make the name of the

character I'm calling it Allen I have no

idea why just came up to me not not very

creative and then you just just drag

those image images into that folder and

when you do that you'll end up with kind

of a library a natural natural growing

library of your character that you can

then use to produce more poses and keep

adding to that now for an improvement on

a method and I might be getting ahead of

myself since I haven't fully explored

the method as it is but I had an idea

and felt compelled to try it and this is

that idea I created a new character and

didn't even do a good job doing so

because as you can see here her nose

here is angular and here is round and

don't get me started on this arm it's

hard to even look at it but let's say

this is ready for rigging I know it's

not really rigging but AI rigging guess

rigging let's let's go with that and

what I I'll do is

repaint select a pose a section of a

pose as I did before but now as you can

see here I prompted for a background in

instead of a action and I kept

everything else I took the image link

that image that I don't use as style

reference or character reference those I

left it here and I hit submit and here's

the result as you can see I avoided

painting around the character which is

the most tedious part of the process and

also I was able to get a background that

actually goes behind the character and

not Blended in with some white or that

it has after this as you might have

guessed what I'll do is repaint the rest

of the

image and for that I'll keep city

background I'll keep everything the same

but I'll remove the CF because the only

character I want in the in the image is

already there and this is what I ended

up with as you can see I have a much

richer much Vivid background and I

avoided that tedious part of painting

around the character so I will be

exploring this method further and see if

I can control it in any way just a few

things you can do either using this

method or the initial method doesn't

matter you can always reframe and by

reframe you can inside reframe you can

use either the reframe tool which is

changing the aspect ratio and

positioning your image on that as aspect

ratio either on Start Center or end or

you can zoom and you can zoom out and as

much as twice and that way you have a

different POV of your character in a

scene also you can edit the prompt when

Zoom zooming or reframing which allows

you to kind of change what will or have

a saying on what will appear on the part

of the image that will be generated and

here's an example of a zoom out of that

image you can have like a different POV

establishing shot things like that and

also here is a reframing as you can see

this is some extra tool that you can use

to make the most of having your

character in different places in

different settings doing different

things and keeping that exact same

character and now for something I

stumbled upon by accident which is a

flaw quote unquote flaw that the mid

Journey team tells us the CF uh command

has which is when you try to get more

than one character on a an image it will

force all the characters to look the

same well that plays out very well for

this method here I was trying to repaint

this area and kind of extend the city

background but I forgot to take the CF

and I ended up with three shots of our

character now this allows me to repaint

any of the sections and have new new

poses of the character or I can even

repaint for example these two and keep

this one and then do the same repainting

this one and this one and keep this one

and reframing it on my my project I can

have the same street the same setting

and my character on different poses on

that set on that setting and here is

what I consider to be a fair shot at

using the CF method with the character

we produced as you can see here I'm

using the images we produced that we

consider to be consistent among

themselves and I fed them all as CF to M

journey I gave it a fair description of

the character and I also used a style

reference because I want them to all I

want to force them all to have the

closest style they can between

themselves and here what it came up with

and as you can see it's not nearly as

consistent as what we were able to

achieve using the method I'm showing you

today so you'll be the judge but for me

this can't be used on the same project

and call this girl the same as this girl

the same as this girl but the ones we

produced can and this is my method for

character consistency using M Journey at

least this far I hope you enjoyed it I

hope you learned something thank you

everyone so much for all the love on

Reddit that pushed me to make a video

like this I never attempt to do one and

I was pretty scared I still am a little

obviously English it's not my first

language but I had a lot of fun and if

you guys want to see more of this I will

be doing more of it please consider

liking and subscribing it would mean the

world to us