Welcome to Kleebz Tech and another video of my
continuing series on Fooocus for Stable Diffusion. This video is basically a continuation of my
last video on face swap and tips for consistent characters. If you haven't watched that, I
recommend watching it first. Don't forget, I have many other videos on Fooocus, from
installation to poses and much more. So, one of the ideas I talked about for
getting a consistent character was having it create a grid of four different angles of
the same person. The challenge was getting the actual angles I wanted consistently. Based
on some comments and ideas I was looking into, I considered using a reference chart to
guide those angles. I looked online and found some references, but
decided to create my own. As you can see here, I changed
this over to a line art setting, 1024x1024. You could do this on speed, which I
probably should have, but what this does is generate a human head facing left or
right. Once you get the angles you want, you can connect all those together. After
generating those angles, I put them together to create this rough draft to use for my final
one. You don't want to try too many angles because when you upscale for face swap, if the quality
isn't high enough, you may not get good results. I've created this one now, and here, what I
would do at this point, once I've gotten those different angles attached and put them all into
one reference sheet I want to use, is click off input image, go into my image prompt, make sure
you have the advanced box checked at the bottom, bring in the rough one I created, switch it
over to PyraCanny, leave those as they are, that's perfectly fine for my purposes here, and
create another one. This will just create a new chart, using the one down here as a reference
to create a bit better one. They're all more consistent. I don't think it probably makes much
of a difference; that's just how I was doing it. At this point, let's say we have
the chart we want, which would be perfectly fine for our purposes. I can go
ahead and cancel this one out down here, also going to uncheck line art, set these
back to where I had them for what I was doing. Okay, so now I have my reference chart finished,
and I can use that to create the different angles for that consistent character. Then I can
go ahead, come up with whatever I want to describe the person, tell it I want a grid of
four images. It's going to use this down here as a reference. Even though I could get that grid
without using that, as my other video showed, you couldn't get consistently different set angles
that you wanted. This way, we're guiding it. Now, I'm going to go ahead here and
generate a character to use for the face swap. I've got everything all set, got this
on PyraCanny, so let's go ahead and generate. I'm going to stop it there because I actually
got good enough results for what I'm going to show. That gives you an idea of how you can
get those different angles now. If we had set these stop at a lower angle, these faces
would have turned more towards the camera, so that's why you want to play around with that
until you get the results that you're looking for. But that'll give you a pretty good idea of how to
get that consistency for the different angles and, obviously, which angle you use is up to
you. You probably only need one profile, left or right, and then a couple of different
ones, and then you can work from there. So now we have our character that we're going
to use, but we want to get the quality higher at this point. So I'm going to go ahead
and go into the upscale and variation. I'm going to drop this down in here now. The key
to remember is when we go into the upscale here, if we do the 1.5 or the 2x, that is going
to change it slightly. It'll vary it subtly, so if you have exactly the face that you're
looking for, or if you use something else as a guide for the face swap to get to this
or something like that and you don't want to change that, you might want to just
do the standard 2x, the fast 2x I mean, because that just does a standard upscale. I'm
going to go ahead and do the upscale 2x on this. Okay, since my system doesn't like to upscale
and record, I'm going to actually use the images I created earlier when I was testing this out.
So, what I ended up with was this image here, which was fully generated and upscaled.
Now, at that point, I split that into four different images. I just opened it in paint,
cropped each section off, and saved them all as separate images. So now I have four different
images that I can use when I want to. Yes, more pictures are useful, and for one reason
alone, is I find a lot of objects. If you have a hat in the picture and you only use one picture
with that hat, that's going to always have a hat pretty much. But if you have multiple pictures and
only one has a hat, then it's less likely going to influence those resulting images, and partially
that depends on the weight and everything else. So at this point, we have our four images
that we have. I'm going to put all these on face swap for now, but I'm going to actually
demonstrate that there are benefits to more than one and just having one. It all depends
on what you're shooting for. Basically, if you're shooting for exactly the angle that
you have, like the straight at the camera, I'd probably just use the one here and I'd get
rid of these other ones because I have no reason to have these other images because they're
going to actually potentially negatively, you know, influence it. And lowering the
weight on them isn't going to help either because that what that does pretty much is it
influences it not only by just not using those as much but also you'll find you'll get less of a
resemblance, I found at least, it seems that way. If I go ahead and I'm just going to say,
let's see, a beautiful woman with red hair, silver blue eyes, oh let's do long. hair. Okay,
so now we'll go ahead and generate. We have this set for all these set on face swap, and we'll go
ahead and generate on quality. Now, the one thing I find when we have the multiple angles in here,
I get more variety when I don't do anything else. When I just put in 'having coffee', if I remove
that and only just have one of these in there, it'll use that. So, if I were to just have the
straight-on one, almost all of them will come out with the straight-on look. If I use one of
these other angles, that'll impact it as well. So that's one thing that I found that really
helps you shape the generation. So, you can use the prompt. I'll show you, for example, here
we'll do. We're going to get rid of these two. So, I'm going to do the same thing, but I'm going
to say 'side' and the thing I've left is down here on the profile view more. It's not exactly a
profile, but it's much closer, and I got to crank up that weight. I should have done that before,
but we're going to do it quite a bit there. Now, I'm going to go ahead and generate again on this,
and we're going to actually use the same seeds for these, and we'll generate. So now we have
shots that are more in tune with that profile, semi-profile shot. I don't know what you want to
call that. But if we have more of a profile shot, I didn't actually have one in one of those,
yeah. So that's when you can use your other slots to bring in, and it doesn't want to do it.
Okay, there we go, and we use Pyracanny. So now we have our face, and we have our pose that we're looking
for, and we're going to go ahead generate. Okay, so you get the idea. Now we have the pose we can
utilize. The problem, a lot of times, you're going to have to work on the eyes. A lot of these things
also, you can adjust the weight. So, I'm not going to go through that and keep generating, but a lot
of times, increasing or decreasing, keep playing around with it. So, if we want to have less of an
influence, you know, we could lower this down a little bit, but you're going to lose some of that
similarity potentially. So as for impainting and face swap, I find to get mixed results. There's a
lot of factors that come into play. The face, what you have, I find when you're generating, it works
pretty well, very well. I actually find I can get very good results. But when you start doing the
in-painting, it's not always the greatest. I'm not a big fan of... Well, let's start with this.
I'll just demonstrate a couple of things now. There's a couple different ways you can
do this, and really, a lot of times, one comes out better than the other. So, I
experiment and usually try both to start with, and then whichever one looks
better, I work from there. We're going to use this image, we're
going to use this one, this face, as the face swap for this. I will paint over
the area. Now, remember, it's going to have to blend things in, so you want to be careful
on how you do this. Sometimes you want to be a little more accurate when it comes to skin
shade because sometimes it will get lines. And then under the advanced, you're going to have
to go into the developer debug mode, and in here, you want to check off 'mixing image prompt' and
'in paint', so you want to have that checked. Now, depending on what you're doing, the method that
you do, you would also need to keep in mind when you—this is what it's normally like when you
do the Imp painting, there's a refiner switch, and what happens is, if you just do regular
imp painting like that set right now without changing anything, the last like approximately
30%, it's going to change the image, the face, so much that it probably isn't going to look like
the same. Sometimes it does and tricked me before, but, so what I find is, you have to go in into
the developer debug mode and also change the 'force overwrite refiner switch' step, and if you
set that up into about 50, 48, you can go ahead and generate and get some images. The problem
is, you're not going to get some of the great ones. You'll get, they'll still work, but, that's
one method of doing it. So you can do it that way. And then the other way is to improve detail, and
then we can go in here, detail, girl face. Now when you do it this way with the improved detail,
you don't need to change the refiner switch, but you do need to have still the mixing image
prompt and the Imp paint set. So we'll go ahead at this point, we have these set to a high
weight, I have the in paint or out paint, and we have that set to the detailed face,
and we're going to go ahead and generate. Well, I'm just going to generate one on that
for now, just to show how it works. Okay, so now we have our final product there. It seemed
to have done a better job than the other one. Now, the other thing you can do at this point,
if we want to, and I'm going to show you a different method now. This can be done with
or without the in-painting. I actually use it, sometimes I'll do the in-painting first like this,
and then I'll go into the upscale and variation. Now in here, we have to change this from the Imp
painting or... well, we don't have to change it, I think you can leave them both on, I don't know.
But mixing image prompt and very upscale now, and so in here, I'm not, I'm going to put it on
this vary, I'm going to drag this down here. So that's the one we just did with the in paint,
we could have done it with the original as well. Hit or miss on which one you can, you can
experiment with all these different methods. This is the face we're going to be using for
the face swap, but now it's going to be under the very upscale. I'm going to get rid of this
one in the in paint, so there's no confusion. And so at this point, we can also try to get a
different expression. As somebody had commented, you can use the variation to try to get one.
So I'm going to go ahead and use a big smile, and I'm going to actually just increase
some weight because I do know one thing, usually you need to increase that weight.
I'm going to lower the stop ad on this to make it a little bit easier,
hopefully, for it can to do that. Okay, so we now have our variation on that
other one, and we can pull that up now. I'm not a huge fan, I still think it's not great.
I would go in with photo editing app and adjust the colors slightly, but it did a pretty good
job with the face swap to get that smile. You do need to reduce the weight a little
bit, so you're going to lose a little bit of the similarities, but everybody's
faces are different when they're smiling, angry, different expressions. So,
that's how the Imp painting works. Like I said, that's hit or miss. I've
gotten some really good results out of this, and then I've got some mediocre results out of
this, and not so great results out of it. But one of the biggest things you want to remember
is, if you're using this straight-on face in the face swap and you're generating with that, the
face is going to always want to be pulling and looking at the camera when you go to generate
anything, no matter what you try to generate. Just keep experimenting, but those are a lot of
the tools that you have at your disposal. The upscale and variation one, I didn't mention in my
last video because I didn't even think about that, but that works, that actually works pretty well.
Sometimes, if there's only one person in there, I will just use that. I won't
bother with the in-painting at all, I'll just skip the in-painting and go
right to the upscale and variation. If you found this video helpful, please do
consider the like button because it does help. Don't forget to check out my other videos
on focus, and any questions or even more tips on getting better results, definitely leave
them in the comments. And have a great day.
