Midjourney: Consistent Characters & Kaiber 3.0

Summary

The video discusses the new character consistency features in mid-Journey, including the release of character references, which eliminates the need for complex prompt formulas or third-party plugins. The host demonstrates how to use these features, including tips on refining character generation and the use of character weight (D-CW) for style adaptation. Additionally, the video explores the latest updates in Kyber's 3.0 model, highlighting its impressive video generation capabilities and the unique, surreal aesthetics it offers.

Takeaways

• 🎨 MidJourney has introduced character references, eliminating the need for complex prompt formulas or third-party tools for consistent character depiction across scenes.
• 👤 The user begins with a character example, the 'man in the blue business suit,' to demonstrate the process of using character references in MidJourney.
• 📸 To utilize character references, one must copy the image URL from the MidJourney website and paste it into the Discord prompt, along with the 'D-CF' command and the aspect ratio.
• 🏙️ The script highlights the process of refining character references through rerolls and adjustments to prompts, aiming for consistency in character portrayal.
• 🖼️ Users can create a model turnaround sheet to reinforce character consistency, which MidJourney can blend together for a cohesive look.
• 🎨 The 'cinematic still' feature allows users to generate images in various styles, such as a James Cameron action film, further enhancing character consistency.
• 🔄 Character weight (D-CW) is a tool that allows users to adjust the emphasis on the original character traits from 1 to 100, enabling style changes while maintaining character identity.
• 🌞 The script demonstrates how character weight can be adjusted to fit different settings, such as changing a cyberpunk character into a photographic style for a summer scene.
• 🎥 Kyber has updated to a 3.0 model, which includes impressive video transformation features and extended video durations of up to 16 seconds.
• 🚀 Kyber's motion slider allows for control over the amount of movement in generated videos, offering a range from no motion to a significant amount.
• 🎬 The script showcases the potential of combining character references with style references, creating powerful combinations for unique and consistent character portrayals.

Q & A

What is the main topic of the video?

The main topic of the video is the introduction and exploration of character consistency in mid-journey, along with tips and tricks for its effective use, and a review of the latest update from Kyber.

What problem does mid-journey solve with its new feature?

Mid-journey solves the problem of maintaining character consistency across different scenes by introducing character references, eliminating the need for complex prompt formulas, third-party plugins, or face swappers.

How does one use character references in mid-journey?

To use character references in mid-journey, one must first have a character image. This image is then copied and pasted as a URL into the 'Imagine' prompt bar on Discord, along with a specific prompt and the 'D-CF' (character reference) command followed by the image URL.

What is the significance of the aspect ratio in mid-journey prompts?

The aspect ratio in mid-journey prompts is significant as it helps control the composition of the generated image. For instance, changing the aspect ratio to 9:16 ensures a full body shot of the character.

How can one improve the consistency of a character in mid-journey?

One can improve the consistency of a character in mid-journey by using multiple images of the same character as character references, which helps reinforce the overall look of the character.

What is a model turnaround sheet and how is it used in character generation?

A model turnaround sheet is a collection of images showing a character from multiple angles (usually 360 degrees). It is used in character generation to reinforce the character's look by providing multiple references, which helps the AI blend the characteristics of the character more accurately.

What is the role of 'character weight' or 'D-CW' in mid-journey?

Character weight or 'D-CW' in mid-journey controls the influence of the character reference on the generated image. It has a scale of 1 to 100, with 100 being an exact match to the original character. Lowering the character weight allows for variations in the character while maintaining some level of consistency.

How does Kyber's 3.0 model improve video generation?

Kyber's 3.0 model improves video generation by allowing for longer video durations of up to 16 seconds, enhancing the facial coherency, and introducing a motion slider that controls the amount of movement in the video.

What is the 'evolve' feature in Kyber and how does it affect the generated content?

The 'evolve' feature in Kyber is a slider that controls the level of warpy, hallucinogenic look in the generated content. Setting it to zero removes this effect, while increasing the slider value introduces more dynamic and surreal movements and transformations.

What is the significance of style references in conjunction with character references?

Style references, when used in conjunction with character references, can enhance the overall aesthetic and atmosphere of the generated images. They allow the AI to pick up on background style details and apply them to the character, creating a more immersive and stylistically consistent output.

What is the limitation of using photographs of real people as character references in mid-journey?

Using photographs of real people as character references in mid-journey is not possible. The AI does not recognize these images as character references, and instead, may generate an image that somewhat resembles the person in the photograph, but it will not maintain the exact likeness.

Outlines

00:00

🎨 Introducing Character Consistency in Mid Journey

The video begins with the introduction of a new feature that allows for the creation of consistent characters in Mid Journey. The host shares tips and tricks to maximize the use of this feature for various creations. They discuss the latest update from Kyber and how it offers a unique and cool experience. The focus then shifts to the newly released character references in Mid Journey, which eliminate the need for complex prompt formulas or third-party plugins. The host demonstrates how to use these references by taking an example character, a man in a blue business suit, and shows the process of rerolling to find the desired character. They also discuss an additional trick for character generation and the use of a model turnaround sheet for reinforcing character look.

05:02

🌟 Character Weight and Style References in Mid Journey

This paragraph delves into the use of character weight (D-CW) in Mid Journey, which allows for the scaling of character consistency from 1 to 100. The host explains how this can be used to change the character's appearance, such as removing heavy armor for a different setting. The discussion then moves to the combination of character references with style references, showcasing how they can be used together to create powerful and varied images. The host provides examples of how different style references, such as those from movies like Mad Max and Tenet, can influence the final output. They also mention the limitations of character references, noting that they cannot be used with photographs of real people.

10:04

🎥 Exciting Updates in Kyber 3.0 and Video Generation

The host shifts focus to the recent updates in Kyber's 3.0 model, highlighting the improvements in the transform feature and the introduction of 3.0 motion. They demonstrate the video generation capabilities, including the ability to create videos up to 16 seconds long with a cinematic preset. The host explores the motion slider, which allows for the addition of movement to the generated videos. They showcase examples of video generation using the new features, commenting on the facial coherency and the unique, surreal quality of Kyber's output. The host emphasizes the importance of maintaining the distinctiveness of video generators and encourages Kyber to keep its unique style.

Keywords

💡Character References

Character References are a feature in the mid-journey platform that allows users to maintain consistency in the appearance of a character across different scenes or images. This is achieved by using a specific image or set of images as a reference, which the AI then uses to generate images of the same character in various contexts. In the video, the man in the blue business suit is used as an example to demonstrate how character references work, showing how different images of him can be used to generate a consistent character in various scenes.

💡Mid-Journey

Mid-Journey refers to a platform or tool that is used for generating images, particularly focusing on creating and maintaining consistent characters across multiple images. The video discusses the release of character references in Mid-Journey, which is a significant update that simplifies the process of generating images with consistent characters without needing complex prompt formulas or third-party plugins.

💡Discord

In the context of the video, Discord is mentioned as a platform where users can copy and paste image URLs to use as character references in Mid-Journey. It serves as a medium for transferring the image link from the website to the AI system for image generation.

💡Aspect Ratio

Aspect Ratio refers to the proportional relationship between the width and height of an image or video frame. In the video, the aspect ratio is mentioned as an important parameter to adjust when generating images in Mid-Journey, as it can influence the composition and the type of shot, such as a full body shot or a close-up.

💡Character Weight (D-CW)

Character Weight, denoted as D-CW in the video, is a parameter that allows users to control the influence of the original character reference image on the generated image. It operates on a scale of 1 to 100, with 100 being an exact match to the reference character. Lowering the character weight allows for variations in the character's appearance, such as changing their outfit while maintaining their facial features.

💡Style References

Style References are used in image generation to capture the aesthetic and visual style of a particular source, such as a movie or a concept art, and apply it to the generated content. This technique allows users to create images that not only have a consistent character but also reflect the visual style of a chosen reference.

💡Kyber

Kyber is another AI-based platform mentioned in the video that specializes in generating images and videos. It has recently been updated to a 3.0 model, which is noted for its impressive video transformation feature and the ability to create videos with significant motion and facial coherency.

💡Cinematic Still

Cinematic Still refers to a technique or setting within AI image generation platforms like Mid-Journey, where a static image is generated with a cinematic or movie-like quality. This often involves capturing a moment in a way that tells a story or conveys a specific mood or atmosphere.

💡Model Turnaround Sheet

A Model Turnaround Sheet is a collection of images that show a character from multiple angles, typically used in the design and animation industries to establish a character's appearance and details from all sides. In the context of the video, it is suggested as a method to reinforce the character's look when using character references in AI image generation platforms.

💡Character Consistency

Character Consistency refers to the ability to maintain a character's visual identity and attributes across different images or scenes. This is crucial in storytelling and content creation, as it helps the audience recognize and connect with the characters. The video discusses techniques and features in AI platforms that aid in achieving character consistency.

Highlights

Mid Journey has released character references, eliminating the need for complex prompt formulas or third-party plugins for consistent character creation.

The character reference feature allows for maintaining consistency across different scenes without the need for bizarre prompt formulas or face swappers.

A simple process of rerolling can help find the desired character, as demonstrated with the man in the blue business suit.

Using Discord and copying image URLs can facilitate the process of character reference on Mid Journey's website.

The aspect ratio adjustment (D-CF) is crucial for achieving full body shots or other specific character views.

Multiple images can be used as character references to reinforce the character's overall look.

An interesting trick is generating a model turnaround sheet to create a variety of poses for character reference without needing multiple images of the character.

Character weight (D-CW) can be adjusted to change the character's appearance in different styles or settings without altering the character's essence.

Style references, when used in conjunction with character references, can enhance the aesthetic and color grade of the generated images.

Mid Journey's character referencing system can blend multiple characters into one, making it a powerful tool for character consistency.

Kyber has updated to a 3.0 model, offering impressive features such as longer video durations and improved motion capabilities.

Kyber's evolve feature, which gives a hallucinogenic look, can now be adjusted to provide more control over the final output.

The motion slider in Kyber 3.0 allows for significant control over the amount of movement in generated videos.

Kyber's video generation maintains facial coherency and character consistency, even with longer video durations.

The character referencing system in Mid Journey is limited to humanoid characters and cannot be used for non-human subjects.

UMES discovered that character referencing can also be applied to non-human subjects, opening up new possibilities for generating logos, icons, or props.

Kyber's 3.0 model introduces beat matching from music, adding another layer of uniqueness to its video generation capabilities.

The video generation in Kyber 3.0 is noted for its unique and surreal quality, setting it apart from other generators.

Kyber's text-to-video options offer additional creative possibilities, though not covered in this transcript.

Transcripts

well gang it's finally here the ability

to create consistent characters in mid

Journey today we're going to take a deep

dive into it plus I've uncovered some

tips and tricks that I think will really

help you maximize your use of it for

your Creations plus I've got a look at

the latest update from kyber I think

you're going to be really impressed with

this it is definitely very cool and

doing something really unique okay let's

dive in kicking off mid Journey have

finally released character references so

we no longer need like really bizarre

prompt formulas or third party plugins

or face swappers to achieve a consistent

character across different scenes to

start off we're obviously going to need

a character so uh I figured we'll take

our old pal the man in the blue business

suit and begin with him so you can

expect obviously a couple of rerolls to

find the character that you're looking

for I do have an additional trick on

this coming up in a second this guy is

definitely I don't know what business

he's in but that does not work

appropriate sir I ended up running

across this guy who I felt was you know

appropriately man in a blue business bus

suie uh I did take him through a subtle

upscale I don't know if that matters or

not but I figured you know can't

necessarily hurt uh from here we're

going to take this over to First Discord

before I show you how to do this on the

website so very briefly just to get the

image off of the website what we'll do

is come over here and hit copy and then

uh come over and copy the image URL on

Discord we issued the prompt photograph

a man in a blue business suit at a City

Cafe drinking coffee in a window seat as

the world rushes past him with an aspect

ratio of 169 and then importantly D- CF

character reference followed by a paste

of the image URL and from here we have

four images of our guy pondering out of

a window probably thinking if that guy

in the Parisian coffee shop and the Sora

video isn't real am I real now to note

these four Images aren't exactly the

character that we had initially

generated and that's okay I kind of

think of this phase as mid Journey

getting to know the character on the

website side what you're going to want

to do is just you know grab your image

and drag him into the Imagine prompt bar

now this is the important part I'm

actually zooming this in because it's a

little bit on the finicky side uh what

you're going to want to do for character

reference is actually you see a little

person icon right there so you're going

to want to make sure that that is

enabled if you leave it on the default

then all that is is an image reference

you're not going to get the same

character you're just going to get the

vibe of the overall image itself so for

this I'm going to swap out the prompt

for a man in a blue business suit

walking through an office lobby and

importantly going to change the aspect

ratio to 9916 that way we should get a

full body shot out of him so we ended up

landing on this guy who looks pretty

close his pants do need to get iron

though uh and now what we can do because

we have essentially our base image we

have our standing image now and then we

have one of our images from the cafe uh

we can actually use this and take all

three of them in and drop them in as

character references as well I like this

one because it was a pretty decent

profile shot as well so we're going to

drop this one in and then finally our

base image uh one thing you can do by

the way is since it's actually sitting

here you can just take this and drag it

right over it's very nice quality of

life uh featured there mid Journey

taking all three of those reference

images and then changing out the promptu

Cinematic still a man in a blue business

suit buying a burrito from a street

vendor uh busy City street yields this

as a result uh yeah that's our guy

buying a burrito off of a street vendor

to note he has had coffee and a street

burrito at this point he is in for a

very long night that said on occasion

mid journey is just going to Mid Journey

for example in our previous burrito

example where uh it rolled up this young

Chad uh kind of like I guess our man in

the blue business suits son taking a

selfie at a taco stand uh clearly this

is all wrong however you can actually

inpaint him out by just selecting the

face of our young whipper snapper here

uh we can then add in-- C ref followed

by the URL of our original character

running that yields us this which is

pretty decent he is starting to drift a

little bit away from the characteristics

of our original character there is a

trick for that we're going to talk about

that in just one second first here's an

interesting trick on generating a

character without needing to generate

multiple images of that character and

that's simply by creating a model

turnaround sheet of that character so

the prompt here is character reference

sheet photographic studio lighting

realistic character reference sheet sort

of reinforcing that prompt T pose uh

futuristic resistance fighter now to

note she is not in a t pose and

additionally she does not look

photorealistic she kind of looks a

little bit more on the Unreal Engine

side but that's okay this will work mid

Journey themselves do state that this

technique does work because the way that

character referencing works is that when

it sees multiple characters it's going

to blend them together so it stands to

reason that if we have three separate

images of the same character that that

would help reinforce the overall look of

that character what's really cool about

CF is the fact that you can change

styles for example in that character

reference sheet she obviously had a very

sort of gamish look to it but running

her through the prompt cinematic still

female resistance fighter battling

robots in a James Cameron action film

Style by Terminator yields us this where

you know she now has the characteristics

of a real person and as you continue to

generate up more images you do build up

a library of that character in various

poses that you can use for for further

character references now as always I

encourage you to continuously reroll

explore swap out your prompts and don't

necessarily expect a home run every

single time um I don't want a cherry

pick here but you know so this was a you

know a four up of one of the uh

character reference prompts uh you can

see here we sort of ended up with a

concept art style uh another sort of

concept art style this one was it was

okay uh and then this one was okay as

well um so if I wanted to continue along

with a cinematic look uh I could grab

both of these two and add them to my

character Library you know it just kind

of occurred to me that she kind of looks

like Jenna Ortega and now I really want

to see like Wednesday Adams versus the

Terminator one command that you will

definitely want to be aware of and use

is character weight or D- CW this has a

scale of 1 to 100 100 being essentially

the exact same character as appears in

your original reference image and then

you can scale that down why would you

want to scale that down well for example

if we wanted to take this image of our

cyberpunk woman with long white hair

standing in a snowy alley and change her

over to a photographic look but then

also set the scene in summer uh you

wouldn't necessarily want her to be in

that level of you know heavy armor so by

taking that image as a character

reference once again we swap over to the

character reference button uh and then

changing the prompt out to photographic

still woman with long white hair and a

black T-shirt intense Gaye City hot

summer day and then D- CW of 50 so half

the amount of the character weight and

run that we end up with the same

character our cyberpunk woman with long

white hair albeit now rendered in a

photographic style standing on some you

know Small Town Street somewhere

definitely heavy summer Vibes but the

bigger note is by taking the character

weight down 250 she's no longer wearing

you know heavy cyberpunk armor which

would look a little bit on the out of

play side and instead is simply wearing

a black T-shirt still has the intense

gaze of a warrior that pretty much says

that you've already lost the battle

suggesting that you know she pay for her

half of the meal so returning back to

our Jenna Ortega sci-fi dystopian epic

you can also use style references in

conjunction with character reference as

well for example taking the overused

Sci-Fi Trope of the mission briefing uh

this is a shot from Prometheus and using

that as a style reference we end up with

this where it definitely has picked up

on all of the background style details

uh you know those sort of blue hued

computer monitors now I will say that

you do run into some problems with this

technique especially if you start

calling out for background characters uh

namely because all of the background

characters then become your main

character I mean granted you could very

easily go through and in paint these

characters out but hey maybe it works to

your advantage like if you're doing a

SQL to uh multiplicity so to use style

references you'll want to end up hitting

this middle button here uh that then

turns it into a style reference once

again character reference style

reference and image reference I always

feel that style references never really

took off I took a whole Deep dive look

at it in a previous video that's linked

down below but now in conjunction with

character references I think that it

becomes a really powerful combination uh

for example taking just like this still

image of uh a shot from Mad Max Fury

Road and utilizing that with our

character reference we end up with with

images like this which definitely have

that sort of aesthetic and color grade

look of that Mad Max shot whereas

utilizing this shot from tenant ends up

giving us this so yeah there's

definitely a lot to explore in here if

you were sleeping on style references

again you could check out that previous

video it is linked down below just as a

note you still can't run photographs of

actual people as character references so

this does not work as like a face

swapper uh for example taking you know

this image of me running it as a

character reference in mid Journey uh

with the prompt man in the blue business

suit uh we ended up with this guy who I

guess you he kind of looks like he kind

of looks like uh my older brother who

doesn't exist uh who I probably owe

money to and who is still disappointed

with my career choices all I can say to

him is who bailed who out of jail in our

20s as an additional note mid Journey

states that CF can only be used for

characters uh namely humanoid characters

that said UMES actually discovered that

you can actually use it uh for different

applications as well uh here taking a CF

of kind of this sprocket uh and running

it through a number of different

iterations so this might be an

interesting back door into generating up

things like logos or icons or maybe even

props character consistency and mid

Journey has long been a hurdle to

overcome and now that it's here along

with the Bevy of other features

available to us uh yeah I'm really

excited to see what you guys end up

creating moving on kyber have recently

updated to a 3.0 model and it's pretty

impressive if you haven't checked in

with kyber in a while I do recommend

swinging by uh the video transform

feature is gotten really good and 3.0

motion well as we'll see in just a

second is very impressive so kicking off

with this mid-journey image of a kind of

robot girl uh and bringing that in

issuing The Prompt robot girl blinking

Smiles uh and then using the Cinematic

preset uh we can hit video settings uh

from here we can select a video duration

up to 16 seconds which is pretty

impressive it's like four times longer

than most of the other standard video

generators uh evolve which is the slider

that I think that kyber is most

associated with it's the thing that kind

of gives everything a weird sort of

warpy hallucinogenic look well what's

interesting is that you can actually

take this down now to zero and then

finally we have a motion slider which we

can take up to 10 for a considerable

amount of motion or you know down to

obviously zero for no motion uh just

make sure the version 3 Model is

selected down here and when we run it we

end up with this which is pretty

impressive the facial coherency stays

pretty solid um it starts to lose it a

little bit around this 7 to 8 second

Mark but I mean you get a pretty good

amount of space before that ends up

happening yeah there is a lot of crazy

movement happening in the background

which you could probably tone down via

the motion slider but I mean for the

most part I think this is probably one

of the longest video Generations that's

managed to stay this C Cent yes yes

outside of Sora but again we do not have

Sora and I'm pretty sure you all caught

the news that we are not getting Sora

anytime soon once again one of our

famous recurring characters Dutch

football player Daniela van Deno dressed

as a pirate uh yeah overall I think it

looks pretty solid she's maintaining

consistency she's not like morphing out

a beard and turning into Jack Sparrow

there is one issue kind of happening

with her hand there there's also no

camera movement controls which is

actually probably helpful in this case

as it helps maintain the overall

permanence and consistency of these

characters taking our man in the blue

business suit and issuing The Prompt

Smiles uh we ended up with this which I

mean that's it's really pretty good

there are some issues obviously with the

flickering but I mean I think that if

you were to take this through a d

Flicker and sharpen it up a bit this

would look really really really solid

but I'll always say when it comes to

surreal and weird like kyber just can't

be beat I mean this is pretty awesome

does it feel a bit like a fever drink

yes it does and that is what makes it

unique in my opinion the last thing that

I want is all of these video generators

to end up having an overall homogenized

look so if you ask me keep it weird

kyber we'll take a bit of a deeper dive

into kyber 3.0 later because there are a

lot of other really interesting aspects

to it including beat matching from music

and uh we didn't even go over the text

to video options here so keep a lookout

for that one in the meantime I thank you

for watching my name is

Tim