Hang tight while we fetch the video data and transcripts. This only takes a moment.
Connecting to YouTube player…
Fetching transcript data…
We’ll display the transcript, summary, and all view options as soon as everything loads.
Next steps
Loading transcript tools…
Create Consistent Characters Using AI | Step-by-Step Tutorial | Curious Refuge | YouTubeToText
YouTube Transcript: Create Consistent Characters Using AI | Step-by-Step Tutorial
Skip watching entire videos - get the full transcript, search for keywords, and copy with one click.
Share:
Video Transcript
Getting consistent characters in your AI
videos is one of the most challenging
aspects of the creative AI process. But
luckily for you, I've gone in and tested
every single AI character creation
workflow, and I have what I think is the
very best way to get consistent
characters that actually look accurate
again and again. So, in this video, I'm
going to break down the entire process
and give you a few helpful tips along
the way. I'm super excited for this one.
Let's hop in. Welcome to Curious Refuge.
So, first things first, we need to
create our base images. And when it
comes to creating the very best and most
cinematic images possible, the best tool
out there is MidJourney. There's all
sorts of advanced features, including
iteration and style reference, that make
MidJourney an incredible tool. and you
can create a lot of images really
quickly. So, the tool is incredible to
use. So, in order to use MidJourney, all
you have to do is go to the Midjourney
website and there's the explore page
which has a ton of really inspiring
photos, but we're going to ignore that
for now. And let's go ahead and go to
our prompt window here. So, for our
prompt, I'm going to keep things pretty
simple. We'll say a cinematic still of a
man in a sci-fi film. Of course, that's
very generic, right? Like you probably
want to go in and prompt for the
specific composition, the lens
qualities, the location, what the
character is wearing, all of those
things. But for our example, I'm going
to keep things really, really simple.
Now, before I hit render, first things
first, we need to bring in our
character. So, for our example, I'm
going to use a picture of me in the
studio here. And I should note that it's
usually best if you have a bunch of
pictures of your subject. You can have
everything from just their face to the
actual clothing that you want to see in
the final result. It really depends on
if you just want the character's face to
be consistent, their clothing, or
basically everything that they're
wearing. It's entirely up to you, and it
will change from project to project. For
this project, I'm going to pretend like
I want that exact same clothing and
character again and again. And so you
can see what I'm wearing here. It's like
a denim jacket. It has a a brown emblem
on the pocket. And then the hair is, you
know, gray and swoopy. And we'll just
ignore the fact that his mustache,
eyebrows, and hair are completely
different colors. So we have our subject
here. And all you have to do inside of
MidJourney is just drag and drop it to
the upload file section. So we're good
to go there. And make sure that you are
in image mode. Right now I'm in video
mode. We'll select image. So, instead of
keeping this image in the image prompt
section, I'm going to drag this over to
omnireence, which is basically like
character reference. And one last thing,
we're going to add in a code to generate
multiple iterations of this prompt. So,
we'll do d-r,
and then you can type in any number you
want all the way up to, I believe, 40.
But for ours, we'll just do five. You
know, that gives us 20 images. It's
really quick. And you do have the
ability to upload style references,
image references, which will create
compositional references that will try
to put the character in the same
location. And you also have the ability
to use mood boards inside of MidJourney
to get consistency when it comes to the
lighting and style and just any other
stylistic elements that you want. again
and again. This video is not going to
cover those things, but you'll find a
link below this video to a few resources
that will show you how to use those
things. Okay, cool. So, let's go ahead
and hit render and see our results.
Okay, and after about 30 seconds, we
have a few images here. Let's take a
look. And you can see that that looks a
lot like me, but it's not exactly me. It
did a pretty good job. But if we want to
get a more consistent character,
especially in the clothing, there's
going to be a few other steps that we
need to take. And we have this shot
here, which kind of looks like me. It's
pretty close, but it's not quite right.
If we're working on a film or
advertising project and we had multiple
shots with this character, there might
be inconsistencies. And then I should
note that totally looks like an AT-AT
mixed with an X-wing, which is uh pretty
cool. We have this shot here. Again, the
beard is completely different. The
clothing is inconsistent. This person is
not me. The clothing is close, but you
get the idea. So, working with
consistent characters inside of
Midjourney can be tricky. You can turn
up the omni reference slider all the way
to 1,00 and you're still going to have
problems getting the character and the
clothing again and again. But that's
okay. I have a workflow that I want to
show with you. So, let's take a look at
three different examples. All of these
images I generated from midjourney using
the method that I just showed you a few
seconds ago. So we have shot number one
here, shot number two, and shot number
three. Let's set the stage. Basically,
let's say that for shot number one, I
want to have the character be, of
course, me. And then I want them to be
wearing the denim jacket from the other
photograph. We'll do the same thing for
shot number two. And for shot number
three, I actually like the clothing that
this character is wearing, but of
course, the character is not exactly me.
They have different facial hair and
their face is just a little off. A lot
of times when you watch video tutorials
showing you how to do consistent
characters, they'll use the most generic
AI looking character that you've never
seen before and it'll generate the
character again and again, but you don't
know that person and ultimately it's not
actually doing a very good job. So, this
method actually will do a good job. So
the next thing that we need to do is hop
over to a really impressive tool called
Idoggram. Ideoggram is an online tool
that can do everything from generating
images to batch generations. There's a
bunch of other editing tools that are
really interesting. But for this
specific example, I'm actually going to
go over to our canvas section here. And
you can see that we have a canvas with
our images. It's actually really easy to
use this canvas. All you do is just drag
and drop the image and it drops it in
there. You can just arrange them however
you want. So, we have our three images.
Now, it's time to swap the face of these
characters. So, what I'm going to do is
I'll select this first one here. We'll
zoom in. And we want to swap just the
face. Again, for this specific image, I
don't want the clothing to change. So,
we're going to go to a little tool on
the left called Magic Fill. I think that
this is one of the most underutilized
and appreciated tools in the entire AI
creative ecosystem. So, let's go ahead
and select our character's head. And
we'll be mindful not to select too much
because, of course, we don't want the
background or the clothing to change
here. And we'll go ahead and select
next. So, you can see we have our magic
fill box here. We're defining that we
want to change the head. And then let's
go ahead and select character here. So
now it's time to go to a new character
and go ahead and click upload. Okay, so
now comes the fun part. Let's go ahead
and select a face that we want to bring
into the generation. So I have a bunch
of different faces here of me looking
around. We'll say we want to bring in
this one here. So we'll go ahead and
bring him into ideoggram. And for our
prompt, you don't have to be hyper
specific, but let's go ahead and direct
what we want our character to be doing.
We'll say, "A man in a sci-fi film looks
to the left." So, really simple. And
then we'll also go ahead and set up a
magic prompt so that it basically goes
in and expands the prompt to be as
specific as possible. It's very helpful.
And we'll go ahead and select magic
fill. Okay, let's take a look at our
results here. We have shot number one. I
think he's looking the wrong way. Shot
number two, basically perfect. It did an
amazing job. That looks like me. It even
has like my cowl hair like pointing up
to the side. Really good. Number three,
not the best. And then number four,
again, it did a really, really good job.
I think number two is just balling. That
is like crazy photorealistic. That's
great. But we're not done yet. So, we'll
go ahead and download that to our
computer. And let's go ahead and do one
more example. Okay. So, we're back here
in the canvas. Let's go ahead and select
this image here. We'll go to magic fill.
Select their lasso. Just grab that guy.
Boop. Just like that. You can select
more if you want or less. It's entirely
up to you. Go ahead and select next. And
again, we need to select our character.
I have found that whenever you are
uploading your reference images, you
could totally just use one image where a
person's looking forward with flat
lighting. That's totally fine. If you
have a wider variety of images so you
can pick and choose the right one with
the right orientation, that's also very
helpful. Again, really flat lighting. If
you have your character standing in
front of a window or a large soft box,
that gives you as much lighting as you
need. It doesn't have to be like really
dynamic or match the lighting of the end
result. So, I'm going to deselect that
character. Let's go over here to the
character button and we will go to new
character. So, for this example, I'm
going to go ahead and select upload and
let's pick a shot of our character where
he's a little more facing forward since
this shot uh is not him looking off like
crazy to the side. So, we'll go ahead
and look through our training data here.
Try to find a good shot. Just kind of normal.
Okay, that's fine. That's like slightly
uh stoned looking, Caleb, but we we'll
take it. So, we'll go ahead and bring
that into ideoggram. And for our prompt,
we'll say, "A man in a sci-fi film looks
up and to the right." And go ahead and
click magic film. And while we wait on
this generation, I just want to let you
know about our AI courses at Curious
Refuge. We train artists at every major
studio and in 172 countries around the
world on the latest AI film making,
advertising, animation, documentary, and
VFX concepts, and we would love to have
you inside of the program. Our artists
go on to land full-time jobs in the
industry, create featurelength films,
and just frankly, they're awesome and
doing amazing things. We would love to
support you on your creative journey.
Okay, cool. Let's take a look at our
images here. So, we have shot number
one. Wow, it put a helmet on me. That's
pretty interesting. Shot number two
looks really, really good. Shot number
three. And then we have shot number
four. I think that like three or four
looks pretty good. Not too shabby. So,
it's kind of hard to pick here between
three and four. They're both pretty
good. I'll just download three and go
ahead and download that to your
computer. Okay. So, now that we've
swapped the face, and it looks pretty
darn good, it's time to swap the
clothing. Now, there are a lot of AI
clothing swap tools out there. And if
they ever allow you to swap clothing,
but you don't have the ability to
actually upload a reference image,
that's not very helpful for a film
making project. However, there are a few
tools out there that do allow you to
contextually edit your photos using
reference imagery and prompts. So, a few
of the most popular are Frames, Nano
Banana, and Seeddream. Now, we've tested
all of these tools, and there are pros
and cons to using them, but ultimately,
I think that Seeddream is the best tool
for doing clothing swaps. whenever you
need to have this conversational
experience to art direct what your
character is actually wearing. And I
think the easiest tool that allows you
to use seedream is freepic.com. So if
you go to freepic.com, go to generate
image and for our image reference, we're
going to bring in our image from the
previous example. So we have this shot
here of me wearing this jacket. Let's go
ahead and drag and drop that into image
reference. And now we need to bring in
an image of our character wearing the
clothing that we want them to wear. So
again, it's a very specific denim
jacket. You'll notice the little patch
on the pocket here. And we'll go ahead
and drag and drop that into image
references. And on the left, make sure
that under model, you're going to
Seedream and then going to Seeddream for
4K. So it's the best quality version of
Seeddream that we have at this time. So
we'll go ahead and select that and then
make sure that the iteration is as high
as possible. At time of recording that
is two, but I like getting as much as I
can because more iteration just gives
you more options. And for our aspect
ratio, we'll do 16x9. So we have our
face swapped image as image number one.
And then image number two here has the
clothing that we want to swap. So we'll
say change the clothes
in at image one. So you see we have this
little box here. So it's going to
reference image number one. We'll say to
match the clothes from image number two.
Keep the original composition
of image number one. And when you're
ready, go ahead and hit generate. Now
you may be wondering why aren't we using
the character feature here inside of
Freepic? The short answer is it's just
the quality is not as good as using
ideoggram. Okay, so I generated four
results. Let's take a look. So, here's
result number one. Again, look at the
clothing. It's exactly the clothing that
I was wearing in the original image. It
even has the overlapping like white
t-shirt that's kind of peeking through
here. Pretty good. Number two, kind of
changed the composition and the lighting
a little bit. Number three, not the
right composition. And then number four,
again, it's looking at the camera. It's
not looking off to the side like we
want. So, I think number one is the one
that we want to use. and we'll go ahead
and download that to our computer. In
most of your creative projects, you
probably would just be able to download
that image. You could upres you could go
into Photoshop and edit it if you wanted
to. But I do think that we can go one
step further and help the character to
be more composited into the final
result. You'll see that in this image,
it almost seems like the character is
lit by completely different lighting
than the ambient lighting inside of the
environment. So, how do we composite
this character to make it look like the
lighting and environment is just more
natural? Well, we're going to use a
feature inside of Photoshop and
specifically the beta version of
Photoshop. So, if you're not already
familiar, go to your Creative Cloud
application if you have a subscription.
I think it's worth it. And go over to
the apps section, click beta, and you'll
see there's a Photoshop beta version.
You can go ahead and download that to
your computer, update it, do whatever
you need to do in order to have this
specific application on your machine.
So, I have Photoshop beta open on my
computer. We'll go ahead and select
open. And let's go ahead and select that
image from Freepick, which again looks
awesome. That is definitely me wearing
the exact same clothes in this cinematic
environment, but I think that we can
composite it a little bit better. So,
first things first, let's go ahead and
select the layer on the right side. Hit
command J to duplicate the layer. I'm
going to turn off the eyeball for the
top layer. And now select the bottom
layer. So, we have only the bottom layer
selected. And let's go ahead and select
the subject. So, it's cool cuz it's an
AI tool. Can basically outline your
subject. You don't have to do that by
hand, which is frankly amazing. And go
ahead and select remove. And that will
remove the character from your image.
So, there you go. We have a blank
background. And I'm going to go ahead
and select that new layer here. So, we
have this eyeball that's basically
covering up the character in the
background. I'll select the bottom
layer. Hit command E to combine those
layers. So, now we have a background
layer here. So, background layer. And we
have the layer on top here, which is the
original character. And let's go ahead
and select the remove background button.
And it removed the background. So we
turn the eyeball on and off. You can see
we have our character isolated from the
background. Now in some use cases, let's
say you have multiple characters or you
need to have just certain elements of
the foreground also brought into that
layer. You may have to go in basically
by hand to edit. But luckily for us in
this example, we're able to use the AI
tools. And I should note that the
character swap feature using ideoggram
also works with multiple characters. You
just select their face and you could
swap them very easily. Okay, cool. So,
the final thing that we're going to do
is select that top layer. We're going to
hit this brand new harmonize button. So,
go ahead and select that. And basically,
Photoshop will go in and automatically
composite the lighting and the color
grading to help the character blend a
little bit more. So, let's take a look
at a few of the final results here.
Again, these are only slight
differences, but I think these small
differences really can help elevate your
work. We have shot number one, number
two, and number three. I think number
three is the best. Again, very slight
difference. And we'll go ahead and hit
file, export, and we'll just quick
export as a PNG to our computer. So,
there you go. That's how to get a
consistent character with clothing
utilizing AI tools. Of course, if you
want to learn the latest creative AI
techniques, be sure to go over to
Curious Refuge, hit that little blue
button in the top right corner, and you
can subscribe to our email newsletter.
We curate the very best creative news
for you each and every week and we'll
even give you a free intro to AI
filmmaking course just for joining our
community. And of course, you can like
and subscribe here on YouTube to get the
latest tutorials and AI news directly
here on the platform. Thank you so much
for watching this video and best of luck
Click on any text or timestamp to jump to that moment in the video
Share:
Most transcripts ready in under 5 seconds
One-Click Copy125+ LanguagesSearch ContentJump to Timestamps
Paste YouTube URL
Enter any YouTube video link to get the full transcript
Transcript Extraction Form
Most transcripts ready in under 5 seconds
Get Our Chrome Extension
Get transcripts instantly without leaving YouTube. Install our Chrome extension for one-click access to any video's transcript directly on the watch page.
Chrome Extension