Welcome to Kleebz Tech and another video on
Fooocus for stable diffusion. In this video, I'm going to be going over some more tips for
consistent character creation that I've been playing around with. This continues with some
of the ideas from my previous two videos that I posted on this subject. I'm also going to
give a couple of other tips that I'd like to pass along as well that I've come across. So,
why don't we just jump right into that now. As we can see here, I've got some consistent
creation going on with characters, and I'm going to show you how to get to what I'm doing here.
Okay, so for the initial setup here, actually, one tip I will give: if you ever find that Fooocus
is hesitating or doesn't want to start, a lot of times this happens whenever you're upscaling.
You'll find this with inpainting sometimes; it needs to upscale the image. And if you
go to your command window and look in there, okay, so sometimes what'll happen is, you're
going to get to a point, if you're watching the command window, it's going to get right
into here, this section here about upscaling. This can happen when you're inpainting,
just upscaling various things like that. I found when I have that happen, sometimes on my
computer, especially when I have a lot of things running, it gets stuck and it can take a while.
But I did discover if I just minimize all my browser windows, including the Fooocus window,
usually it'll kick in within 10 to 15 seconds, it'll start upscaling at that point, and
then I can maximize my window again and continue where I left off. So, that's
one of my tips that I've come across. Onto my ideas for accomplishing what I have
here. So, this, I'm going to use the speed settings initially. A lot of this, I'm actually
not going to go through every single step. I'll show you what I did. I've already created
some of these things. I don't want to make you sit here and watch me create them, so I
would put this on the speed setting for now. Depending on what you're looking
for for poses, this is to get the poses that you want. As you can see in
this grid, we got four different poses, like I did here with comic characters.
But if you're doing something like that, you may be able to get away with even more poses
with upscaling and splitting them. But I'm not going to jump into all that right now. I'm just
going to show you how I did it with these four. I don't need the image prompt or anything.
For style, I'm going to uncheck all of these, and I just want the line art. I'm going
to leave the model as the Juggernaut, which is the standard. I'm not going
to change anything in there. I'll go ahead and start making the
models that I want to use. Now, what I have found is to do this, you just
want a very basic structure for the image prompt to follow. So, just like I did in my last
video with the headshots, I can generate, using a 3D mesh character, a human form, male,
standing, white background; alter that as you see fit. For this purpose, though, we just
want the basic pose. I don't want any extra stuff that's attached or any clothing to show or
anything like that because that'll influence the final image. So, you may have to generate several
different ones to get the results that you want. Okay, so for the purpose of my demonstration,
this gives you an idea. You'll be generating different images. I prefer the ones that are
just very simple and basic. Sometimes it will put clothing on them or things like that, but that's
usually I find can influence the final image. So, I find it's best to try to get the most basic
ones that you can get. Then, you come up with multiple different poses that you're looking for.
So, I took multiple different ones, pasted them together. I use GIMP for that. I find bringing
one in and expanding the canvas allows me to just drop the other ones in, and then I can get four
different ones. The benefit of that is I'm keeping the same aspect ratio, because that's one thing
that you want to do. Stick with certain aspect ratios; if you start mixing and matching them,
you're going to start getting some odd results. Now, once I've done that, in here, I'll
go under the image prompt. I'm going to bring that one in because these are all rough
ones; the models are slightly different. So, I find that I just want to use that as a
reference. I'll go ahead and drop that in, and then I'm going to tell it here, we'll drop
this one as just an image prompt as a reference, so it'll know what I want it to look like. And
then I'm going to do a grid of four different poses, 3D mesh character, human form. And I've
got the same aspect ratio, whatever that was, because these, yes, this image is twice as
large, but it's still the same aspect ratio. So, I can use that to influence my final. And
this, as I said, is just on the image prompt. Now, you may need to adjust these weights
to get what you're looking for. Still, keep this on speed. The style is still line
art because I'm still generating the same one. You could just use that original one that
I've used, but the proportions are going to be different. I find by regenerating
another full grid based upon that, it does much better at getting those
proportions the same for each model. So, when you go to generate the actual
character, those will come out much better. Okay, now once you get the results that
you want, like I said, you're looking for just the consistency between each model, as
long as that's sticking with what you want. I'm going to go with this last one, just
for what we're doing at this point. Now, I can remove the image prompt here,
and I'll bring this one down here. So, that's now going to be my reference that
I'm going to be using. Now, I'm going to drop this down a bit, and these you'll have
to adjust to get the results that you want. Now, at this point, I'm going to change it
from line art because I've got my poses, I've got what I'm looking for. Now, I want to go with
my final, and I'm going to do a comic character, basically. For this one, you could do this with
photo-quality stuff, but it's going to be a little harder because you're going to have to do a lot
of upscaling, face swapping, things like that, and you may lose some consistency between those.
I find, especially for these sorts of images, comics, or anything else like that, you
can usually get much better results. The other thing to keep in mind when you're
creating these things is simplicity is your friend. If you make an outfit that is
very complex or anything like that, it's going to be a lot harder for you to get
more consistent results as you upscale and everything else. So, simplicity is usually what
I find to be the best when it comes to this. I go ahead; this is going to be my poses. Now, I
can at this point also use a face if I wanted to, although I'll usually use that again later
to improve the detail. So, I went ahead, and I generated this earlier. So, this is actually
going to be the face swap that I'm going to use; that'll be the face that we're going to use
on this character. And I could also put in, if I wanted to, the clothing to
influence it, things like that. But I'm not going to do that for this one.
I'm just going to create, let's see here, let's go with a man in a blue suit and a
black tie in different poses, full body, white background. And that pretty much is all I need
for this. I can touch some things up afterwards. So, let's say we have this point. I'm going to
increase the quality, although since I'm doing this under the comic settings, I find a lot of
times you don't necessarily need the quality setting. But I'm going to use it for this one.
Everything else, I've unchecked everything else because you So, I'm just going to go ahead and
skip that because this will not get everything perfect every time. Sometimes you're going to
have to try multiple times, but I find usually what happens is it will consistently create the
same character, same clothing in every single one. Sometimes it might be slightly off, which you
can fix with inpainting if you want. This usually works pretty well in my experience. I have done
it with photo models, styles, etc., and it works, but you're going to have a much harder time when
it comes to the face and hands. Then you have to do a lot of inpainting to improve those. And as
you upscale the images, if the details of the clothing and everything are more intricate, you
can start to lose some of those details as you increase the resolution, and then your consistency
will start to fall apart. So, like I said, this seems to work much better. I find for things
like comics, 3D characters, things like that, photorealism is going to be a bit more work to
get that consistency. It can work. I've played around with it, and it does a pretty decent job;
it's just not as easy to do as I'm showing here. Okay, so we have our sets of images here.
They're all pretty consistent. Actually, this one picked up what I was going for; this one
is holding a cell phone and talking on a phone, so I won't have to edit that one. I had to do
that earlier, I had to change it using inpainting, and this one, it added a briefcase. So,
I'm actually going to use that last one. Now, at this point, what I would not do is,
I'd find the image that I want and I can open it. How you do this is up to you. I use Paint
for this, only because I find it to be easier, and then I can go ahead and pick
the one that I want to improve upon. So, let's say, okay, so we have our two.
I go through, I do all four of those, or however many images you had.
I've now split it into four, so the resolution is actually lower. So, I'm
going to go ahead, I'll get rid of this. I'm going to leave the face swap in here,
now I'm going to upscale at this point. Now, this is the part that sometimes can take
a few tries because you have multiple ways of doing this. I prefer to use the upscale 2x. It
does change the image a bit, but I found the quality is better than just doing a standard
upscale. The standard upscale, the fast 2x, will just upscale it, but you're not going to
get more details or anything. So, I'm going to leave this on the same style, we're going to
leave it on the same resolution, that's fine. Oh, the other thing I want to do, I find
that works pretty well, is if you go into the describe tab here, and then you drop your image
in here. We're going to switch this over to art, anime, describe this into one prompt. I find
I can drop this in, it'll describe the prompt, I can make sure everything looks fine in here.
Now, I'm going to go back and upscale. I also want to use the face swap. You could do this
later with inpainting and that sort of thing; that's up to you. I probably still
will, but I find through this step, this process, it works pretty well to keep
using the face swap. But in order to do that, to use it for the upscale and variation,
you do need to go into your advanced tab, developer debug, under control settings,
mix image prompt, and vary upscale. So now, when we generate this, it'll upscale it and use
that face that we have set in the face swap. Okay, so I'm actually going to stop
it right there. I've got a variation upscale that I want. It actually looks,
if I look at these, everything's pretty consistent with the original one. So, I will
go ahead and use that one at this point. So, I can uncheck that, and we'll go back into the
image. I could bring it into the inpainting, and I could fix anything that I wanted to
at this point. If the face wasn't exactly what I was looking for, I could use the
face swap. So, let's go ahead and do that. So, I've gone ahead, I got the face swap set;
that'll work for our purposes. I'm going to set that pretty high. Now, I'm going
to go into our inpaint, and there are a couple of different options here. We can go
in, we do have to go into the developer debug, we have to check off mix inpaint or image prompt
and inpaint. Now, there are two options here we can do this. If you're going to use just
regular inpainting like I have chosen here, I haven't changed it down below, you do need
to go in and increase the forced overwrite refiner switch up to 50 or so because once it
switches the refiner and the regular inpainting, it will lose that likeness, so you want to
increase that higher. I find in most cases, though, I don't even do that. I just leave
it as normal, I go down here, improve detail, detailed man's face, and then I'll go
ahead and I'm just going to generate again. Okay, so we have our finished product on
this one. I'm going to actually go ahead and just do the other one
as well. Let's, um, I'm not going to go through every single one. This is just to demonstrate
how this works. Obviously, you want to wait until you get the image that you prefer. But
once you've got that image that you prefer, then I'll go ahead and do the same thing I did
before. If I'm not satisfied with this face, I got, put the face swap in there, and I'm
going to go ahead and mask that as well. Now, this one, I'm actually going to change a
little bit. I'm going to paint off the hand, and I am going to set it for inpaint. Now, in
here, since I'm going to do the regular inpaint, I'm going to change the refiner switch. I'm
going to set that up to a higher amount, and I'm going to uncheck that. We'll do the,
well, let's put in, talking on a cell phone, add that to the prompt, and now we're
just going to go ahead and generate. We do have, like I said, we've checked everything,
the inpaint, so it's mixed, our face swap is set, so let's go ahead and generate that. That last one
looks pretty good. So, at this point, if I wasn't satisfied with the face, I could just go ahead,
just do the face, now the way I did it before. And I'll do in here, now the only
thing we need to do with this is, okay, I've already left that
overwritten, so we're going to turn that down now. We're going to go
to just the face, like we did before. So, let me pull up what we have, just to get an
idea of the results. And, really, I didn't spend much time on these. You know, you could get
much better results with a little more work, but that should give you a pretty good idea of
some of my concepts that I've come up with. You can combine this with the heads, with the whole
body, there's a whole bunch of combinations. I'm going to continue to keep playing with some
of these ideas and just keep coming up with more options on what you're looking for.
Obviously, this isn't going to do well if you need a whole bunch of different images,
but it's really good if you only need a few images of the same character in different
poses, is one way of getting to that point. Okay, well, I think that sums up what I wanted to
cover today. Hopefully, a lot of that made sense, and hopefully, this video will give you some
more tools to create consistent characters. If you found this video helpful, please
consider hitting the like button because it does help. Don't forget to check out my other
videos on Fooocus as well. Any questions or tips on getting better results, please leave
them in the comments. And have a great day.