VisoMaster - Face Swap - Замена лиц
Hello everyone, Oleg and the StableDif channel are with you. Today we’ll talk about replacing faces using the VisoMaster application. VisoMaster is an application that does Face Swap at a very high level, that is, replacing faces in images and videos. There is no censorship, and the speed of work is very high. On the 4080 and 4090 cards you can do Face Swap in real time. Just like you see on this screen now. VisoMaster has a lot of different settings, but this application is very convenient to use. It's intuitive. If you are not familiar with face replacement technology,
then don’t worry, with this application you will definitely be able to change faces in both photos and videos. Video memory requirements according to the developer are from 8 GB. Works only on Nvidia video cards. It works faster than the same reactor in ComfyUI or Automatic. And importantly, there are automatic masks that can cut out hands, hair, glasses and other objects from the surface of the face. The installation is very simple, like any Windows application, but it will download about 30 GB of various models and files. As a result, you get a portable assembly. You can naturally transfer this folder
with the assembly from disk to disk, and this will not break the assembly. But I have prepared an archive of installed VisoMaster for you. It is compressed to 13 GB. Accordingly, you simply download this archive for 13 GB, and not 30, and simply unpack it where you need it. As a result, you get the exact same build as if you had simply installed and downloaded 30 GB. A link to this archive will be attached to Boosty, and to the installation files it will be
attached under this video. If you want to learn how to use the Automatic and Forge web interfaces, then I invite you to my courses on the website stabledif.ru. The link will be in the description under this video. There are also administrator contacts there. If you still have any questions, be sure to ask them to the administrator. I look forward to seeing everyone at the courses. Now let's continue.
So we will have two installation options. The first installation option is to simply unpack my archive like this. And the second unpacking option is installation from the developer’s website. I 'll show you both options. Let's take a look at the first option first. Download my archive. It will be added to Boosty. Place it in the folder where you want to see this version of VisoMaster. We upload it and copy it. After this archive has been copied from us, right-click and select unpack using the 7Zip archiver. If you do not have a 7Zip archiver,
download it from the official website 7zip.org. Click here to unpack. We choose where to unpack it all. I'm unpacking directly into the same folder, so I just click OK. We wait until everything is unpacked. It's very fast, literally takes less than a minute. And after that we will be able to use VisoMaster. That's it, VisoMaster is unpacked. Here it is, this folder. We go to this folder and double-click on the file start portable,
namely portable, and not just start. And we are launching a portable version of VisoMaster. Here we can replace faces in both images and videos. Let me now immediately show you the installation of the official source, and then we’ll start understanding this application. Now about installing the official source. The link will be included in the description below this video. Well, of course, it will also be included on Boosty. You click and go to this resource. Below you will find the releases. Here you only need to download
this exe file VisoMaster setup. Click on it with the left mouse button. A download window will open . Click save. I've already downloaded it. You will have a file like this with this exe icon. Double-click on it. The first thing you see is probably security information. Then click Next in this window, that is, the usual installation like any
other Windows application. We choose where we will install it. I'll select the same test folder here . I click here okay. And here I’ll just substitute number one, so that it’s a different folder, a different installation. I click Next. A window like this opens for us. Here we must select the version of our CUDA that we have installed on Windows. To do this, click here
, in the search, write cmd here. You get this command line application. Click on it. And here we insert this command, which is called NVCC space dash dash version. Press Enter and find out what version of CUDA you have installed on your system. Look, if you have 12.6 installed like mine, for example, CUDA does not need to be rearranged, everything will work for you anyway. It is important to simply choose the right generation. If you have 11.8, it shouldn't be 11.6, but it could be 11.9. And if you choose 12.4, you should not have
less than 12.4. But maybe 12.5, maybe 12.6 there. That is, you can choose this option. I select 12.4 here and click Next. Look, if you don’t have CUDA installed at all, how can you install it? How to download it? Write directly to CUDA and choose, for example, 12.4. Press
Enter. In your Google search, you will definitely find this version of CUDA Tool Kit 12.4. Click on this link and go to the Nvidia page. Click here Windows. Surely you have Windows. Click 64 and Windows 10 or 11. I have the eleventh. Click Network or Local. Local - this means that the entire 3.5 GB package will be downloaded at once. And Network - this means that a small installation file like this will be downloaded, and it will load all the necessary libraries that it will need. I usually download Network, it weighs less, you can save it there somewhere, then
you don’t have to go to the site, just run it. We simply close this entire window and return back to our installer. Click it means 12.4. Download Model and Updates. This checkbox will mean that all models and all updates will be downloaded from you. If you don’t install it, then you will have to download it manually somehow . It's inconvenient. Therefore, click Next with this checkbox checked so that it
loads everything itself. And click Next and click install here. After I click install, the installation will begin and all these models will be downloaded. This will all take approximately 30 GB. The developer writes that there should be at least 50 on the disk. This is how it should be, because perhaps
something will be expanded there, some caches will be downloaded there. Therefore, 50 GB is the minimum. The archive that I showed you weighs only 13.4 GB. This is significantly less than 30 GB. It will naturally download faster. If your Internet speed is not very high, then it would be more preferable to download this archive and unpack it. Well, in general it will just be faster. Click Install here and VisoMaster will be installed. That's it, we now have either one or the second version of VisoMaster installed. We can
launch it using start portable. In both the first and second options, you need to click start portable to launch it. If you want to update it, you can always update both versions using update portable. You launch it, you get an update. In fact, VisoMaster is updated quite often, so the update will be current. I've
already launched it. Therefore, this window appears: continue the workspace or start over. I click continue. OK. In general, this is what VisoMaster is like. Let's look at an example image first. I select the image here. How to do this? Yes , nothing will happen here . Either we click in the center of this window, or we drag some images directly here, we can drag them directly like this. Or we can click here on this folder and
select a folder with images or videos. Again, here both the image and the video are all mixed up. That is, we select it, and everything is loaded there. Now about this window here. Here we must load the faces we will replace with. Again, click either this icon, or drag and drop the faces that we want to place here right here. See, I've uploaded it. Look, this is the image I uploaded here, but here we just got one face. This application automatically trims everything, that is, it sprinkles only
the face, identifies it and sprinkles it, and leaves only it, so that everything is replaced as efficiently as possible. Now here at the top we select what we will change. For example, this is the image. Here just below the image we have some buttons. Click find, that is, find a face in this
image or video. I'll show you next. This is the face we have. On the left we select, What will we replace it with? For example, this is the face, and click Swap Face. Swap Face means face transfer. That's it, the face has been replaced. It’s such poor quality, but don’t be alarmed, we’ll get it right now. That is, this is the face we got. We click on another face, it is replaced by another, a third, there and so on. Now about the quality. Look, there are different tabs here, there are a lot of different options. Let's just go straight through it in order
. The first thing you need is common. Basic settings here. Face Restore, that is, restoration of faces. What's the problem? Yes, look, we have models for replacing faces like this. That is, there are a lot of models. In ComfyUI you probably didn’t have so many models, but in reality Inswapper 128 it still works with the highest quality. In some cases, where you need to get some kind of style, you can try Shost Face, you can try Instyle Swap. By the way, it also works very well in some cases, but in most
cases with video and so on, basically this Inswapper 128 works with the highest quality. 128 - this means 128 pixels by 128 pixels. That is, the resolution is very low, so you see, the image is clear, but the face is so blurry. Here we have options to select higher detail. You see, the detail is higher. We choose even higher detail or even higher. Look, the detail really gets better. But nevertheless, if we look at the eyes,
they are still not very clear, not particularly correct. But nevertheless, you can still choose 128, of course, yes, but you can also choose 256, 384. Indeed, probably the result will be a little better. Next, go to common and here select Face Restore. Let's press it. Select
a model here. There are different models for Face Restore here. We are familiar with GFPGAN, if you use ComfyUI, that is, these are the usual models for restoring faces. But there are these GFPGANs, by the way, they are also in ComfyUI. They are available in different resolutions. And 512, for example, install it like this. That is, it depends on a specific image or a specific video. That is, you need to select, that is, which model is best for you. You see, for example, now I’ll click GFPGAN 512. The result, it seems to me, is much better. Again, in combination
with this resolution. Here, in general, select it to suit your needs. There is also a blend option, that is, mixing. And accordingly, we can weaken blend from 100% a little bit like this, get something average, a little softer. That is, you see here, at 100%
you get such a straight-up sharp face, but you can weaken it a little and we’ll get something more or less like this, more acceptable. Maybe. And again, there is the power of this model; again, you can play around here. There is a second restorer. This is basically the same thing, but it results in an additional repeat pass. But again, here we can lower the blend there, but in this case it is not necessary. I'll just turn it off here
. And also on this tab we also have this option enable face Expression Restore. This is something similar to the Life portrait that we have in Automatic in ComfyUI. Here we have the opportunity to control the lips, eyes, in general, restore them and change their position, say, that is, give some kind of facial expression. The only thing that needs to be taken into account is that this option puts a lot of load on the video card. Accordingly, you won’t be able to do this in real-time video with this option on any video card, that is, there will be slowdowns. But if you just sort of calculate the video, then this can be used. I didn’t notice any special effect from this , but in some cases it may also be useful. Let me turn this off,
and let's go to this tab now. On this tab we can restore colors there, that is, this is color correction. Face Editor, that is, color correction of the face, eyes, and so on, lips. In general, separately. And Face Swap. Here we have a lot of different tools.
Let's now move on to the video so I can just show you the useful tools that are here. Well, it’s clear that Face Ice can probably be shown right away, that is, this is eye restoration. But keep in mind that this parameter is, of course, customizable, but nevertheless it changes a person’s face a little, so you can only use it carefully and make some small changes with it. Likewise, Restore Mouth is the restoration of lips.
Here, too, there is a slight opening of the lips, some teeth appear there. In general, this slightly changes the character himself, of course, so be careful too. Color is an automatic color correction of the replaced face so that it fits better into our image. A very useful
option. Color adjustment - again, this is a manual color correction; you can do some kind of color correction if you don’t like the automatic color correction. You can change it manually. Well, final blend is simply mixing what you had initially with what you got. But it seems to me that this is unnecessary, only the faces initially and afterwards were very similar, then in this case something might work out normally. Let's now move on to the video to see other useful options. Let's choose this video. Now I have uploaded this video. Look, here we can scroll through and select
the most suitable frame to capture the face. For example, let's take where the face is a little straighter. Well, maybe that's how it is. Let's take this face. Find face. We click on this frame, and accordingly it finds the face. Now we select a face here, for example this face, or for example this face. Let's take this here. Or maybe this, yes, so that it could be something else. Look. And now we can just go and start watching the video. You see, we are watching in real
time, but my stream is currently being recorded, so it slows down a little. But by the way, resolution also influences, of course. If we put here, for example, at least 128, and here we also put, for example, a regular one, then it will be much faster. That is,
on the video it will be like this for the video. You see, here GFPGAN is as close as possible to a regular one. And just 28 Inswapper fits great here. So it works great here. That's all. We can already observe the result in real time. Well, let's find a frame where our face is slightly blocked by our fingers. I specifically chose this video here. Look, it’s
normal here, yes, but it should be like this. Look, here I am flipping through it now. That is, the fingers should be completely visible. Here we have a spoiled video, and fingers are simply broken, cut off, let’s say. This is very bad. But there are such options as automatic masks. There are two of them here
. And in some cases, one option works well. In some cases, another option works better. Let's enable the first option one by one. And look, now our fingers have appeared. Now nothing spoils here anymore. That is, the fingers are cut out like this with a mask and they remain untouched. Everything inside this white mask is replaced. That is, the face is determined and changed. This is not our face, but our hand. This is what the model determines. And accordingly, there are no problems. Let's take another look. It has virtually no effect on speed. It works very quickly. This option basically does the same thing. You can enable both. Well, this is just unnecessary resource
consumption . It's just that sometimes this option works better in some cases. In some cases, this one here. That is, we do this and get this result. Well, let's now take a video with two characters. For example, here's a video. Let's look a little bit
where the faces are. We'll be a little closer. Well, like this for example. Click here find face, that is, we find both faces. You see, both the first person and the second were found. We choose the first person. Here we select, for example, this character. Replacement face.
Let's choose the second person here. We choose a second character, for example. This is what we'll choose. And we replace the face here. Let's see what we ended up with. It turned out like this. And here is the result. Look, in some cases it works worse, so you can put
X Mask here, and maybe it will be a little better. But this is the result that comes out in the end. Again, you can experiment here. Let's just stop some frame so that nothing slows down, and select Inswapper here, for example. Let's leave it again. Let's choose 512 here. Just like that. And let’s choose here as we chose GFPGAN 512. That is, it turns out like
this. In principle, the quality here is quite high, but it worked out. This is a little better, but with these parameters, of course, I won’t be able to scroll this picture directly in real time. There will be brakes like this. Therefore, in real time, with these settings, your card most likely will not work, unless you have a professional video card. Now let's see how it works with a video camera to do all this live. Click this
checkmark button. Look, now I clicked this checkbox. Here I have a window with my webcam. I click on it too. And I appeared here in this field. I'm motionless here , as if my camera isn't moving, but if I press play here, you can see that I've started to move here. That is, now I am mobile here. And if, again, I leave it like this, then I’m just stopping the picture so that nothing seems to slow it down, yes. We can naturally stop at one picture and press find. Again, the face is determined. Look, here
I choose, for example, this face. And I click on it, and let's click Play here. And now you can watch in real time how my face is replaced. If I move further away, the result becomes much better. If I'm closer, of course some
artifacts are visible there. But again there are different settings here. We can put there, for example, 385. I hope I won’t slow down now, but there should be a little more detail. And we can put here, for example, GFPGAN, at least 256. This is how everything is processed in real time when recording a stream. But maybe this way, yes, it’s better. That is, maybe ComfyUI in some case will eat a little, yes, it seems to me that after all. Or GFPGAN, here in general you need to select. That is, depending on the lighting. And lighting has a very strong effect,
that is, it’s like they’re illuminated. How do you set it all up here with the light? Here is GFPGAN, it basically works quickly. These are the settings. Again, you can cut out the grease there. For example, look, I’m now putting my hand to my face, yes, my hand is not cut off, that is, there are no problems. For example, I even cover my face like this, there are no problems. And if I close it now with both hands , you see, such artifacts appear, and my face pops up there. That is, it’s not ideal, but yes, it works. Again, it depends on the lighting, but nevertheless, that is, it basically
functions in real time. You can get into trouble with your friends, let’s put it this way. I hope you won't do anything stupid with this. Here. That's it, let's try it simple. By the way, you can switch between faces in real time. There’s this, this face, there’s even a woman’s face there, for example. In general, any face can be switched. Again, the face should suit you, of course, but look, here’s an example. Kerifs, yes, just like that. Well, if I pull my chin down, it turns out better
. If you make such a round face, then it won’t look like it. Here's Pushkin, please. That is, Pushkin is simply from a painting. This is DiCaprio. That is, this is what he looks like. Here in the picture he is shot a little from the side. And if this option is on the side, it looks better. Look, here you can also mix faces. For example, I click on DiCaprio, hold down Ctrl on the keyboard
and click on Pushkin. Now you get a mix of Pushkin and DiCaprio. Well, let's try Pushkin and Kean Reeves. It turns out like this. In general, you can do cool things in general, but in reality you can upload several images of one person, and you will get a better result than with one image. That is, I didn’t include several faces here, but you can try different options like this. Look, there are a lot of settings here everywhere. And I probably didn’t show you the main option here, yes, when you work with video, for example,
like this, yes, click here, here. You see, everything is set up very quickly. Swap, that is, the face has changed, and clear means clear it here. Here. And click here start recording, that is, recording. But I don’t have a folder selected where this is being written. Therefore, go to settings. And in settings we Here we select the folder where we will save everything. For example, on the desktop, I select select folder and again click record, record. And you see, I am calculating and saving this video. That is, it will now be calculated to the end. You can see right away, everything is in real time.
It all works very quickly. That is, despite the fact that I’m still broadcasting this to you on video, that is, the video card is loaded at the same time, in the end I got this file. Let's open this file. This is the output result. That is, you can definitely try it, no matter what powerful or not powerful video card you have. I think even on mediocre video cards it will work quite quickly and comfortably. In settings we have a lot of different settings here. There you can set the FPS you want to ultimately see. There is a choice of
TENZOR RT, TENZOR RT Engine or just a processor. That is, it affects the speed of rendering. The processor is very slow. CUDA is just any Nvidia video card, and TENZOR RT is accelerated generation, but there the preliminary calculation still happens directly for a specific video, so well, this may not be relevant in all cases. But if your video is very long, very large, then perhaps TENZOR RT or RT Engine will be relevant. That is, you will have a preliminary rendering
of some section of this large video, and then the generation will go much faster. But in these cases, it’s probably worth trying to choose it. In other cases, it’s just that this calculation in advance will be longer than when you calculate just such a short video there with 194 frames. Well, other settings here are already specific. Here you can select various manipulations with rotation, heads there, there, and so on, to adjust somehow more precisely. Here you can choose the resolution of the video camera, there is Full HD, not Full HD or just HD, depending on the power of your hardware, which can handle such a resolution, yes, maybe there in real time or not in real time. Here you can already adjust to a
lower resolution. Maybe it will work for you in real time, but a higher one won’t . Again, the number of frames per second that is calculated. 23 will naturally be faster than 30 or 60. Therefore, again, there are different settings. Do you need some kind of smoothness or, on the contrary, do you need speed. That is, here you can already choose. And another interesting option.
There is a virtual camera here. Why is this necessary? Look, you may have your own video camera, yes, but naturally it won’t do this face transfer, yes. For example, would you like to change this somewhere in Skype or some other video communication application? You need a virtual camera in order to broadcast from it as if from this application. To do this, we turn on this checkbox here , select where we will broadcast, whether it’s in OBS or some other application, and select various settings here. These settings here are selected for
online, say, broadcasting for faster speed. In this case, you will have a virtual video camera called VisoMaster on your Windows system. And you can simply select it in Skype or another application, and broadcasting will start from it, and not with your native camera. That is, from your native camera it goes here to the application, and from here it goes to Skype or other other applications. Friends, in general, this is how the review turned out. Really interesting
application. It weighs 30 GB when expanded. It will weigh like that. But again, if you don’t need some models there, you can probably delete them. But if you update VisoMaster, they will definitely be downloaded again. Therefore, there’s probably no point in deleting it.
30 GB weighs quite a lot. But on the other hand, if you need some task, you can unzip the archive, do your business there, change it there and delete it again. And this archive can simply be left to store only 13 GB. This is quite convenient in my opinion. Thanks for watching. If you were interested in this video, be sure to like it, go to my resources, go to my Boosty, there are many different other videos that you will also definitely like. Come to my Telegram channel Stable Diffusion and my big Telegram chat Stable Diffusion.
There is a very large community on neural networks in various areas. You can chat with colleagues with similar interests. With that said, have a successful generation and good mood. Bye everyone.
2025-04-17 13:54
