Это видео недоступно.
Сожалеем об этом.
FaceFusion: The Definitive Deep Dive & Walkthrough - Everything You Always Wanted to Know About...
HTML-код
- Опубликовано: 16 авг 2024
- I tried to be as in depth as possible and not leaving any stones unturned. If after this video, you don't know how to use FaceFusion, I'll give you your money back.
There is a ton of information in this video and a lot that you may or may not already know. So I'll try to make the chapters as useful as possible so you can easily skip to the parts that most interest you.
Clean Install Video: • FaceFusion 2.0.0 Clean...
==Chapters==
00:00 - Intro
00:42 - Quick Software Upgrade
03:20 - Absolute Basics
05:00 - Execution Providers
05:55 - Max Memory
06:10 - Output Path
07:11 - Options
08:20 - Preview & Preview Frame
08:54 - Trim Frames
09:42 - FSM-Reference
11:23 - FSM-One
11:50 - Face Analyzers
13:27 - FSM-Many
13:53 - Face Detector Model
15:25 - Face Swapper Models
17:11 - Face Enhancer Models
19:00 - Frame Enhancer
19:46 - Face Debugger
23:43 - Face Mask Box
25:25 - Face Mask Occlusion
26:20 - Face Mask Region
27:06 - Things I Missed
27:39 - Outro
![New AI FaceSwap tool ! Rope[Crystal] step by step installation !!](http://i.ytimg.com/vi/JFmzsinsaFM/mqdefault.jpg)
00:08 Ensure you have the latest version of FaceFusion installed
02:31 Setting up FaceFusion for different platforms
05:08 Adjusting thread count and memory for optimal rendering speed
07:19 Settings and options explanation
09:33 FaceFusion allows easy trim frame adjustments and face swapping with reference options
11:54 Explanation of different analyzers in FaceFusion tool
14:27 Adjusting Face Detector Score and Frame Processors Face Swapper Models
16:54 Face Fusion offers seven model options for enhancing faces.
18:59 Overview of FaceFusion features
21:08 FaceFusion allows users to pick a reference picture and manipulate key points for tracking.
23:26 Fine-tuning face selection and masking in FaceFusion
25:25 FaceFusion introduces smarter masking tools
I watched this yesterday and was working in FF today. I'm doing lip sync with face enhancer, but the enhancer is ruining the eye blink I already have in the face. Then I thought I recalled you saying something about masking and came back to watch that section again and I now have a good lip sync, without ruining my eye blink! Thanks!
Thanks so much for this video. I have been searching for explanations for what these settings and sliders actually do. Came up empty handed, that is until today when I saw a Reddit post on FaceFusion, and your video was referenced as a great source of information. So glad that I did. Thanks for your efforts!
oh mate. this has been one of the best tuts ive ever seen for ai stuff. thanks heaps
You're very welcome. Thanks for watching!
Great job. Really great, thank you. It's pretty annoying to generate test frames for every combination of every model/setting. Even when a preview is lightning fast. Some of the modules are ambiguously named. It's really nice to have an overview. You've definitely saved me a ton of time. FF gives pretty amazing results right off the bat without touching anything, but dialed in... holy crap. Really, really cuts down on the post processing. A few masks here and there, some interpolation. Amazing. Thanks for your work.
This is absolutely the best video explaining the features of face fusion. Thank you very much sir!
"I'm the dude, playing the dude, disguised as another dude." hilarious!
when i Download the output after face debugging the footage downloads with face trackers still intact? how to remove the tracker defore downloading the output
You need to render it without face_debugger turned on. The debugger is just a tool to see what is going on "behind the scenes". Any settings you change while using it will still be kept when you turn it off.
TRhank you so much you saved me a lot of time✌
First of all, I just want to say thank you for all that you share with us.
I truly appreciate it.
I am impressed with the quality of the final products when using faceswap, but I work mainly with the lip-sync option. My issue is that I attempt to add my own character voices and impressions to various characters which are anamorphic, but Facefusion doesn't recognize many of the faces that I create with Fooocus. Is it possible to manipulate the facial recognition points to get the images to work, or are there any settings I can tweak to have a successful lip-sync? Thank you for taking the time to read this.
I assume you meant "animated", as "anamorphic" is shooting widescreen on a 4:3 format. That being said, yeah, FaceFusion and almost all face swapping apps will not work well or at all with animation. It all depends on the detail of the faces. If they have distinct face landmarks. Sometimes, it will work ok with 3d animations. I got some "ok" results with How to Train Your Dragon, but not great. 2D animation, that definitely won't work. And because you're doing more of the lip-sync side, that's probably going to be more difficult. And there are no settings at all for the lip sync model. Anyway, good luck and thanks for watching!
My goodness, I meant to say "anthropomorphic"!
Thanks for responding, Shady.
I was hoping that ai had brought more to the table for working with non-human faces in the lip-sync arena.
Years ago, Reallusion made a program named Crazy Talk, which allowed you to lip-sync any face imaginable by letting you manually manipulate all of the nodal points of the face, but the actual lip-sync itself needed a lot of manual tweaking, too.
Who knows? Maybe next week someone will make it happen.
Thanks again for replying.
@@alanevans3447 yeah, animated or anthropomorphic, still basically the same answer, as I'm sure you knew. But yes, it seems I've seen similar functioning apps in the past for stuff like that, but nothing I specifically remember.
What to do if i want deepfake someone eating because when the person start putting something in mouth it's just become mess up or the mouth not open
It's not always going to work. But I would try turning on "occlusion" instead of "box". If there is too much going on, a clean swap just might not be possible. You also have the option to turn on "region" and turn off the mouth options. That could help, but that means you would retain the original mouth. You could kind of do the same thing if you stayed in "box" and instead adjusted the bottom of the mask so it doesn't affect the bottom of the face.
Hi! I have a problem with the program not following mouth and eye movements, as well as retaining the same expression despite the source video changing (smiling, talking, etc). The final output thus becomes very stiff. What should I do? Thanks in advance.
Without actually seeing the footage you are using and all of your settings, there really isn't anything I can do. My recommendation would be to go on FaceFusion's Discord (which should be linked in the notes) and ask for help there where you can post screenshots, etc.
Also if your face doesnt change in the preview box before rendering is that okay? 😊
Also answered in discord, but too much to type here now. But simply, no, it's not okay. 😉
Can you share what you know on Face Mask Types, thanks.
Not going to go into great detail here, but box is for a basic mask that allows you to pull in the side of it to mask things off in a rectangle. Occlusion is used to automatically occlude anything that goes in front of the face (hand, beer bottle, an elephant, whatever). Region - whatever you turn off will keep the original of that item from swapping. These can be mixed up. I talk a bit more about these in my FaceFusion Basics 05 video about the Face Debigger. I'd recommend checking that out.
amazing thanks so much. would be amazing to get one like this for the newest 2.5.2
Most of the information is the same. But I do plan to eventually make videos of the updates.
Great Video.
Thank you for making this video. I wanted to ask like when the view is side of the face the swap image will drop out . What should I look into to reduce this or is a bug? This was also happening in roop unleashed. thanks
Sadly, it's a limitation of the model. There isn't anything that can be done to fix this. Other than someone create a better model. Roop used the same model. And it's the best model that's available currently.
I have an AMD gpu, how do i turn on Cuda? im on pinokio, its been taking literal days for some video's to render
Cuda is for Nvidia GPUs only. There is an accelerator specifically for AMD, but my brain isn't working right now. They'd know for sure on the discord channel and know how to install it. The discord is linked in the description.
@@shadyendeavor Oh ok, Thanks!
Thx, great one!
W Video
I'm going to take the "W"!!!
thanks !
#3 : you mean to import multiple images of the same person with different angles to enhance to improve the result ?
Correct. Different angles (not extreme as profiles don't work) and lighting can help the resulting image look more accurate to the target.
Good day, when I press the middle tab, which is the masking option next to the box on the screen, an error appears on the videos, I cannot mask.
How do you handle situations when the target turns profile and the mask slips or bugs for a frame or two, any tips on settings to help?
Thank you for awesome vid! My exuction providers say Azure, and CPU. I do not have coreml. Does Azure also speed up processing or do I need coreml? Im on an Acer running off windows
I believe you already got your answer in the Discord ;) But for those not there, coreml is for mac only. Azure is not worth using, but it's for MS Cloud Computing and I don't think it's ever really been used. Otherwise, your specific system issue was answered in Discord.
What benefit is there to facemask debugging? Is it just to make sure that it selects the correct face?
At 19:46 there is the chapter marker for Face Debugger that goes through all of the features, benefits and uses for it, which include the following three chapters that show the 3 different types of face masks. But yes, it has everything to do with the face masks - which face, how many faces, if it will select during every frame based on the score, etc.
How is the privacy, I want to use this for fun for my family, but is there a chance that the movies can be leaked when using facefusion?
Completely private. Once installed and running and whatever models you are going to use are installed, this can be run without internet connection if you were that unsure. But there is no internet activity at all while using this app, other than when downloading new models when you switch and you can see all of that live in your terminal window.
VERY GOOD! THANK YOU
Glad you liked it. Thanks for watching!
Could we add more than one images as source. Is it working with only one image?
27:24 - I post at this time stamp this exact feature, so yes, you can. I'm not sure what you're asking about "working with only one image?" though. As yes, of course it works with only one image as that's the basic functionality and how I show it being used through the entire video.
Hello, after downloading the video, the face was downloaded in a red frame.
It sounds like you left "Face Debugger" turned on when you started the render. Just make sure you turn it off before rendering out.
@@shadyendeavor Thank you, there is another problem. As soon as I press occlusion masking, which is the most important masking feature, the videos disappear and an error message appears. As soon as I press the middle masking button, an error appears.
how can you get rid of the nsfw filter?
Sorry, you're going to have to try to figure that out on your own.
Thanks
Can we drag drop 5 face photos and swap face each photo to a specific target face ? If it is a group of people in the video
No, at the moment it isn't possible like that. You need to run through one face at a time. Then run the output video for the next.
@@shadyendeavor thx
Huh. I missed the other video with this.
Thanks a lot for this great guide! After some hickups on installing I have faceswap up and running on a M1 16gb pro. When using just faceswap I get an output (I use a 15 sec flick) in 30 seconds. But when I use face-enhancer to booth for a smoother look the timer keeps ticking and the rendering process just never stops. Is there any adjustment I need to make in the settings or do I ask my laptop something it's not capable of?
Without seeing what you are actually doing, it's very difficult to tell you what the problem is or if there even is one. When you say the time keeps ticking, are you only watching the output box in the FaceFusion interface? If that's all you're watching, that might be the perceived problem. I almost always watch the progress in Terminal that actually tells you which process it is on and what frames, etc. and the reason it might seem like it's not doing anything is because face_enhancer takes literally 10-25 times longer than the basic face_swapper.
So with your 15 second clip - let's say that's 30fps and 1080p - that would make the clip 450 frames - on my very similarly equipped 16Gb M1, that would take about 10 minutes.
IF that isn't the issue and I just wasted a bunch of typing 😉 it could be a few other things that I'm not really the one to diagnose. The best thing to do would be to go to the MacOS Troubleshooting channel on the FaceFusion Discord channel, which I should have linked in the video description.
Man, facefusion giving me errors out the butt. I have been trying to get it to work, and it is quite frustrating, especially when the installation instructions are can be understood by a simpleton.
2 options, well, 3 actually.
1 - go on their discord and ask for help with whatever specific errors you might be getting.
2 - Use Pinokio instead to install and run FaceFusion. I have a video for that too. ruclips.net/video/LyNxnHjU32I/видео.html
3 - Try it out on RunDiffusion. They have a free trial and then the pricing is actually super cheap as it's based on your actual usage and so you can spend as much or as little as you want. I make a ton of swaps each week and it costs me about $2 each time with an extreme amount of time saved compared to doing it on my computer. I also have a video for that. ruclips.net/video/lzLRCSWq7Aw/видео.html
Hopefully one of these options works for you. Good luck!
This thing is grrbage... it fails on every video file I feed it.