IMPORTANT - READ THESE UPDATES 2 things were left out of the original installation documents. so be sure to do the following: During stage 1, once you've done the conda installation, close your terminal or cmd window and open up a new one and continue where you left off. If this wasn't done, sometimes this can result in an error: CondaError: Run 'conda init' before 'conda activate For stage 5, the command use to just say "python install.py". Now it says "python install.py --onnxruntime {default, ...}" And this can be very confusing as neither of those will actually work. What the new version actually means is for you to type in your accelerator type where it says "{default, ...}". So the following are all of the possible commands that you are supposed to run based on your system setup. Windows & Linux with NVidia Graphics (CUDA and/or TensorRT) python install.py --onnxruntime cuda Mac (Apple Silicon and Intel) python install.py --onnxruntime default Windows with AMD Graphics (DirectML) python install.py --onnxruntime directml Linux with AMD Graphics (ROCm) python install.py --onnxruntime rocm Windows & Linux with Intel Graphics (OpenVINO) python install.py --onnxruntime openvino
@@shadyendeavor It's the first time I've felt confident about using the terminal for AI, so thanks. Pinokio for my AMD/Nvidia linux PC didn't set up Facefusion that well. So I figured I needed to learn.
@@crippsverse I completely get you on this. I was never a Terminal person. It wasn't until AI became a thing that I "forced" myself to use it. I'd say "learn", but I don't think I have learned it yet. But whenever I had to install something with command line, there was always the worry of missing something and screwing up my computer (which now I know is pretty difficult to do). But now that I've probably done hundreds of installations since and have messed around with a few other things, I've wanted to make sure that whoever did watch my vids, that they had full confidence in what they were doing by laying out exactly what everything was and what it was doing. I hated the videos where people would say, "do this, this and this" and never give you any idea of what each step was actually doing. Anyway, I'm rambling now, but again, thanks for watching and I appreciate the comment.
One step was conveniently left off the installation documentation that isn't always necessary, but may be a problem for some people. During stage 1, once you've done the conda installation, close your terminal or cmd window and open up a new one and continue where you left off. If this wasn't done, sometimes this can result in following error the error: CondaError: Run 'conda init' before 'conda activate
@@ghostatschoolmemories3232 99.9% of the time, the default inswapper is the best - especially when used with pixel boost. I don't really ever do still images, but I think one of them might be slightly better for them.
@@shadyendeavor Well the default (inswapper_128_fp16) work fine but the face detector (yoloface) glitch specially when a person face show half or he or she see above once their eyes are out of sight the detector can't able to keep them up though I really appreciate your work and it's really amazing how far the face fusion has came my suggestion if the tracking can be improve similar with any 3d face tracking
@@ghostatschoolmemories3232 Until better models are introduced, there is nothing that can be done about any shots that are past the profile or not showing enough of the face. It's just a limitation of the current models. This most likely won't fix these issues, but I always recommend choosing "many" for the detector. This will add an unnoticeable amount of time to a swap, but this uses all of the detectors instead of just one and they all have their strengths.
I'm learning so much from your videos. Thanks for time you take in putting together such great content and for the straight forward manner in which you explain things.Quick question. When using FF for lip sync, face enhancement, and frame enhancement for a video, have you noticed that doing each of these three steps separately reduces rendering time in comparison to checking all three boxes and doing them all at once? Thank you again.
Hi, everything in the install was going smoothly until I got to the end of Stage 2. After doing conda activate facefusion, I get "CondaError: Run 'conda init' before 'conda activate'". This happened on both 2019 iMac as well as M1 MacStudio. What to do about this?
I'm hoping this is the issue, so try this. But one thing that should be done that was conveniently left out of the documentation is that once you've installed conda, that you should restart your terminal session by either closing the window you were working in or just completely quitting and re-opening terminal and continuing where you left off.
@@mr.8oom you mean for when say 3.0.1 comes out? Simple, be in your FF directory and conda environment, then: git pull python install.py python facefusion.py run
@@8561 the intro background? Do you mean the background that's throughout the entire video of the girl in the tank top with my name on it? I'd have to look it up, but I know it's pretty complicated and uses style references in MidJourney.
@@8561 The things I have in the prompt that might help you get what you want are "funk art portrait", "Gil elvgren and jimmy turrell style", and "risograph printed portraiture on artist’s paper". The rest is just adding the text and describing the girl. Well, and the SREF #, which was 3061430968 for this one.
I don't have a video that specifically talks about it, though I do have a video where using that feature is a very large part of what the video is about. Check out the following video about morphing in FaceFusion that is all completely done by using multiple source images. th-cam.com/video/Rz_Zfqn9d2U/w-d-xo.html
Thanks for the video tutorial. I got it up and running. Wondering now why under execution providers CUDA is not showing? I'm only seeing a CPU option right now.
were you not given the option when you did the step of "python install.py"? It SHOULD give you the option to choose from whatever accelerator you installed in step 3. To try it again, make sure you're in your directory and have your conda environment activate and then install the accelerator: conda install conda-forge::cuda-runtime=12.4.1 conda-forge::cudnn=9.2.1.18 Then do the install again with that one specifically chosen: python install.py --onnxruntime cuda And then do the conda deactivate and activate again: conda deactivate conda activate facefusion If it's still not showing up, you're going to need to ask for more help on the discord. Good luck!
@@shadyendeavor Thanks for taking the time to respond. I realize now the silly mistake I did during installation. Your troubleshoot helped me fix the issue. Cheers!
Did I miss something or didn't you explain WHERE exactly the installation is going to be put on the drive? I collect all my installation in a special sub folder.
You didn't miss anything. I never said where it installs. But with the regular install, it goes directly into your home folder. If you want to have it install and run from somewhere else, you're either going to need to know how to do that yourself or ask for help on the Discord.
hi, thank you for sharing! The upgraded version is amazing! Btw, I have a question about a face leaving or entering the frame, and I wonder if you can advise me on how to avoid "glitches"? Or do you have a video that talks about it?
It's a limitation of the models and there really isn't anything to be done about it other than choosing footage for your targets that doesn't have these situations in it. If the detector models don't see enough of a face, it can't recognize it AS a face. And if it's not detected, the swapping model doesn't know that there is something to swap.
You can't with just FaceFusion. The outputs are always based on your target video's length. So if you have a still image, that's all you are going to get. You will have to use a third party software to create an actual video of your still image. Something like Davinci Resolve can easily do this. I'm sure there are other options out there that are much simpler to use if you aren't familiar with editing software.
I've used Microsoft Clipchamp for that exact same purpose. It's a very simple video editor in comparison to the others I have. What I do is put a single image of a face into Clipchamp, or any other video editor for that matter, and just keep on adding the same image repetitively, say for about a 60 second video. I then go back and figure where the 1/3 and 2/3 points are on the total timeline. For the images that are between these two points, I flip those frames horizontally to give some realism to what otherwise would be a very boring video of a single image. I also do a fade transition at the these two points, just to add some flair. Once I have the completed 60 second or so video rendered, I save it to my PC and then go back and put that 60 second video into Clipchamp and basically do all that again to match the video length to my audio track length I want to use for lip synching. It's certainly not the ideal solution, but it does a good job for basic lip synching. Hope that helps.
@@iheartjetaviation2764 since we all have ffmpeg installed for FF, there's a much easier way if you're comfortable with command line. Just use the following command: ffmpeg -loop 1 -i /Volumes/Prometheus/image2videotest.jpg -vf "pad=iw:ih+mod(ih\,2)" -c:v libx264 -t 10 -pix_fmt yuv420p /Volumes/Prometheus/image2videotest.mp4 Obviously, change the input and output files to the correct directories and names. And then the "-t 10" is for how many seconds you want the video to be. Does it almost instantly. The weird thing in the middle is modifying the image so that the height is divisible by 2, as you can't create a 264 or 265 if it isn't. Weird, but necessary.
You're not dumb, but also, this video wasn't about the setup or usage of FF. I just put in the quick setup that I used in order to test and show that the installation was successful. I'm not exactly knowledgeable on the subject, but for me (on a Mac), I have it setup so that it's on an external drive named Prometheus and within a few subdirectories. So my output path looks like this: /Volumes/Prometheus/Downloads/Videos/FFOutput/ If you simply wanted it in your regular Downloads folder on a Mac (if you haven't changed the location), then it would simply be: /Users/username/Downloads/ If you're using any directory names with 2 words with a space, that always makes things more complicated. I'm not sure if this works on Windows and Linux, but one thing that is very simple to do is to have a new terminal window open and literally drag the folder/directory of the location you want onto the command line and it should populate with the path. Including if you have multi-named directories. Hopefully this helps and thanks for watching!
Thanks for this video. Could you please explain what the advantages of using the Windows Installer version of FaceFusion as opposed to using it in Pinokio or via the Chrome link within Pinokio. By the way, I have tried both within Pinokio as well as via the Chrome link (my GPU is the Invidia RTX 3060) and so no reduction in rendering times and the same good quality of the rendered result. Does the windows installer speed up rendering times. Is it worth the money?
For the most part, not really any. Pinokio literally goes through the same exact steps as the separate Windows and Mac installers (and the exact same as this tutorial, in fact). Pinokio has the added benefit of having multiple AI apps all easily available together. I'm not sure what you mean by the "chrome link". I'm assuming you're talking about the link that will open it up in your default browser instead of running it within Pinokio. If that's the case, then again, no real benefit either way. Other than Pinokio is better as it's not spyware riddled like Chrome and you should be using Brave instead. :P The installer has no benefit other than hopefully simplifying the install process for those not comfortable with following a tutorial like this one or even comfortable enough to have first installed Pinokio.
@@shadyendeavor I apologize for the delay in thanking you for this explanation. I now can rest easier knowing that I'm not really missing out on anything by sticking to my current method. Love your videos!
I could be wrong, but I believe it should be fine. I'm pretty sure Catalina and earlier are the ones that it can't install on because of software dependencies.
Watching this video I have a bunch of questions. First the execution providers i noticed you deselected cpu and only selected the coreml. I only have cpu, cuda and tensorrt as my options and I atleast make sure cuda is sorking but I make sure to select both CPU and CUDA because I think 2 is better than 1 right? usually when in doing a video face swap i have it running and its taxing my CPU and my GPU both at up to 100% usage. I have Intel Core i7 , 64 GB RAM GPU:NVIDIA RTX 4070 running off a Samsung 980 PRO with heatsink NVME drive connected straight to the motherboard (my root drive) I do have 7 HDD's inside my machine but that root drive is my fastest and big enough at 2 TB's. I also usually push the execution thread count to somehing between 9-11 mostly at 9 and the execution queue count usually at 2 or somethign 3. I also never change the face selector from reference because I never am using anything that has more than one face in it really. ANd I seen that for the source you through in a list of files im assume are all different images of the face you are using, but why do you put multiple fast files in, do you see a difference? And the detector model you picked many as Ive always just picked yoloface just because it seemed like it always was the best as far as speed and quality mix. I also would add occlusion and region to the face mask types and for images i go as far as processing face_swapper, face_enhancer & frame_enhancer. I also pick the 1024x1024 face swapper pixel boost & inswapper_128_fp16 , for face enhancer i think i wouold normally pick comedformer_plus_plus or transformer whatever i know i might be mixing those two up. and ultrasharpx4 for the frame enhancer. but unless its important i dont usually use frame enhacer at all for a video because it slows everything down to nearly a stop from what it seems like. I dont know anything about what i am doing other than trial and error with every ai tool ive ever started using like auto1111, comfyui, facefusion, rope, forge and even others so im not saying im making the smart choices. All i tend to have are chatgpt plus for help in learning any of it and sometimes watching videos like yours but reading tutorials tend to be almost impossible because i dont tend to retain much information i read and its my slowest form of learning anything. Just wanted to know how wrong people think i am.
First, realize that this was an install video. Not a "how to use facefusion" video. I did a quick setup to see if and show that it was working. As I say in the video, I'm on a Mac, so I use CoreML and not the other options. As far as between CUDA and TensorRT, I also say in the video as to which one you should choose or go ask for more information on the Discord as I don't know enough about it. As for picking CPU, you only choose that if the other options are not available to you. Picking CPU along with the other options does nothing as FF will only use one. Yes, you will see your CPU being used either way as there are tasks that only run on the CPU. But you choose the accelerator that works best for when there is a process that CAN use it. So no, choosing 2 isn't better than one in this situation. Otherwise, all of the questions about Face Selectors etc are easily answered by just watching my videos. I have a Basics video that is specific to the Reference selector and another that is specific to the One and Many selector. Reference - th-cam.com/video/l8D5Of1tLM8/w-d-xo.html One & Many - th-cam.com/video/e6hMCa4tg1Y/w-d-xo.html For the detector, Many uses all of the options instead of just one of them. So it has a higher chance of detecting the faces. It might be a bit slower, but nothing noticeable. I also go over most of the other settings in my Deep Dive video. Deep DIve - th-cam.com/video/nvPbdHVF12c/w-d-xo.html 1024x1024 is overkill is most all situations. 512x512 is much better to use and takes literally 1/4 of the time to process. I have a lot of videos on FaceFusion. Best thing would be to just check out my entire playlist on them if you really want to learn as much as you can about FF. Playlist - th-cam.com/play/PLMwq9QXPN0iDReOut5vKekXT4V6wGTET8.html
@@shadyendeavor oh Yes i know what you are talking about in your video. I just do always pick them both because my amount of time does tend to be a bit shorter with cpu and cuda selected but ill do more experementing. I have been using face fusion for a while too, over a year but like i said just using it trial and error learning. I have been watching alot of your videos the last 2 or 3 days and ill keep it going, thanks for all the help!
Also I was wondering why do you have videos in other languages on this channel and not make a separate channel in each language? It would benefit you more I would think as you can eventually get to monetizing from multiple locations doing that. I just know a few other people that do it that way and it tends to upset people that subscribe and get another language video in their feed.
@@Skettalee if you looked, you'd see that on every dubbed video there is a pinned post stating in that language that all dubbed videos are being moved to a separate channel. www.youtube.com/@shadyendeavordubs At this point, there are only 8 videos left on my main channel and I'm moving them over one a day.
CondaError: Run 'conda init' before 'conda activate' I have this error at the end of stage 2 someone can help me I dunno what to do! I tried to activate face fusion but impossible
Actually, not a dumb question. You have 3 options. 1. (I don't recommend) Just do the install and it SHOULD overwrite everything, as it usually would. However, 3.0 is a big upgrade and a lot of changes were made and even the models need to be changed out. 2. Go to your home directory (which is where facefusion should be installed) and just deleted the facefusion directory/folder and everything inside and then do a clean install. 3. Same as 2, but instead, rename the directory/folder to anything else (even just adding a letter to it will work) and then do a clean install. The new install will ignore the old directory that you changed the name of. Doing this will now keep a backup of your old install, just in case something is wrong with your new install (which there shouldn't be). Once you know everything is working well on your new install, feel free to delete the old one. Either way, deleting the directory is the same thing as "uninstalling" facefusion. There are no other files to worry about.
Assisting your algo: I use a Mac normally, I have a hackintosh, about to build a newer one (both with triple boot). I’m investigating a used Epyc system with a plan to use 4 old Nvidia server GPU or 4 newer AMD gpu. ROCm seems to be stable enough for ai gen
I honestly didn't realize that that was a possibility to run rocm on a hackinstosh. That's impressive if you're able to get that to work inline with FaceFusion. Though FaceFusion will not work with multiple GPUs. It can only send data to one. However, you can have multiple instances of FaceFusion running to do multiple jobs at the same time and have each going to a separate GPU. Then I'd just be worried about the CPU being the bottleneck. Or the heat. Let me know how it goes. Would love to know if it works.
@@shadyendeavor I miscommunicated, I wasn’t clear 🙈. The hackintosh is strictly i7 and built in video. I’m not patient enough to tinker with an AMD hackintosh and researching will take time away from using the tool The Epyc will be Linus and window solely for LLM and stable diffusion
@@peterxyz3541 ah, gotcha, that makes sense. Would have loved that to be a solution for crappy Mac AI. But definitely understand not wanting to take the time to figure that hurdle out. If it's even possible. Either way, good luck with the new system.
it's skipping and not overwriting all target frames when the image is small. So if the target has, let say a mustache, the resulting video will show mustache popping off and on. it also doesn't replace head looking up or down. the face flickers to the previous face. Need an option to force replace all frames.
sorry to bother you but when i run "python install.py --onnxruntime {default, ...}" i keep getting this error "zsh: parse error near `}'" and now im stuck
you're not putting it in correctly. I even typed it out and put it on the screen 08:50 Whenever you see curly brackets in command line, that means those are the options within to choose from and enter - not to actually duplicate exactly what it shows.
@@shadyendeavor Hi, sorry..I have the same issue. But the result is zsh: command not found: python. I'm on mac. Could you please help me with this error? Thank you
Thank you for this video I have a problem in step 7, I get this error when excusing the command: Traceback (most recent call last): File "/Users/jeremy/facefusion/facefusion.py", line 7, in from facefusion import core File "/Users/jeremy/facefusion/facefusion/core.py", line 6, in import numpy ModuleNotFoundError: No module named 'numpy' Thanks for the help
be sure you have your conda environment activated. On the line where you input the run command, does it say "(base)" or "(facefusion)"? If it says base, that means you are not in your conda environment, so type in: conda activate facefusion It should then say "(facefusion)" at the start of the line, then try to run. Otherwise, something else along the way was missed or skipped. But most likely it has something to do with the conda environment.
@@jeremyheyob It still, most likely, has something to do with the conda environment, regardless of it being activated. It's beyond my scope though. I've just seen that most people with this problem on the discord are able to fix it by going through the steps again, which either means a step was missed or input incorrectly, or it means that a step failed somehow.
@@taobaofy8500 I'm on an M1 Mac, so most likely, it runs slower for me than you. However, if you want to run it online, RunDiffusion is currently the only option.
Pretty much anything above a potato. But seriously, it's just as long as your system is new enough to actually be able to install the different components. As long as they can install, the app should run. It's just a question of how slowly.
Using simple git pull upgrade, will it work going from 2.6 to 3.0? Also... do I just enter 'default' when choosing my onnxruntime? (sorry... not a coder)
"Hello, on Windows, when I try to run the Conda widget, it doesn't work, and this message appears: An existing package has already been installed. Attempting to upgrade the installed package... Failed to upgrade the package using Winget. Use the method provided by the publisher to upgrade this package
Sorry, but that's definitely a question that's out of my league. Especially since I'm on Mac. Ivan or Happyfunball should be able to help you one the FF Discord. Link in the description.
Hey, I just came across your videos after searching for a solution to my problems, your videos are really insightful, thank you for making them 🙏🏻. I'm going thru a problem , when i face swap, some lines around the swapped face are visible in output video. Can you please tell me how to fix it ? I'm really new to all this. Thank you, and appreciate you for your time.
If you're referring to when it kind of looks like a ghost frame around the sides of the face, I get that a lot too. Try switching between using "occlusion" and "box", or just one of them if you have both turned on. Hopefully that should fix it. Otherwise, depending on the footage you're using, it might be possible to use the masking to cut it off. You'll have to turn on the debugger in order to see the masks and then try to "shave" off the ghosting. Pain in the butt, but not much else to really try. Just make sure you turn off the debugger when you render out the video/image.
@shadyendeavor Hi, thanks a lot brother, changing face mask type to occulsion helped. It worked perfectly. Thanks a lot again, really appreciate it. 🙌🏻
All seemed to go ok, but upon running, I see no execution provider other than my CPU. It swaps images fine but videos aren't working (presumably because it takes forever).
My thinking is that I needed to use the --onnxruntime cuda command, instead of default. Now I don't know what to do. I'd like to uninstall and start from scratch but then, I don't know how to uninstall either. I'm going to risk just renaming the facefusion folder and doing another from the start.
Pleased to say, that all worked. Rename the original folder, reinstalled from scratch and used the 'cuda' option in the install instead of 'default'. All working now :)
@@shadyendeavor Thanks for replies. Sadly, upon closing and opening again, it again doesn't have cuda option for the accelerator :( It did after the 2nd install, but now it's gone again.
I have no idea. There are way too many factors that could be the problem. Your best option is to contact the actual developers on their discord channel which is linked in the description. And please, stop using caps.
There are so many reasons why that could be "normal" and not. It depends on many different settings. First, what kind of computer are you on, Mac or PC? Are you set to only CPU? Did you install the accelerator correctly as stated in the pinned comment? Are you only set to Face Swapper? Do you have Face Enhancer turned on? Do you have Frame Enhancer or any other models turned on? Do you have Pixel Boost turned up?
🙀..no Premiere...???? this is outrageous 😾 [Kitty launches into tirade begins clawing all furniture in home 😾] . FaceFusion 3.0 Epic Trailor 👉 th-cam.com/video/2QzTj6o-pv4/w-d-xo.html
@@shadyendeavor What are the dependencies??? It run and prozess but show no output. On win 11 I dont get it to run. It is apain I tried so much cost me days.
![I Replaced ALL my ADOBE APPS with these [free or cheaper] Alternatives!](http://i.ytimg.com/vi/5EfqHg49kMk/mqdefault.jpg)
IMPORTANT - READ THESE UPDATES
2 things were left out of the original installation documents. so be sure to do the following:
During stage 1, once you've done the conda installation, close your terminal or cmd window and open up a new one and continue where you left off.
If this wasn't done, sometimes this can result in an error:
CondaError: Run 'conda init' before 'conda activate
For stage 5, the command use to just say "python install.py". Now it says "python install.py --onnxruntime {default, ...}" And this can be very confusing as neither of those will actually work. What the new version actually means is for you to type in your accelerator type where it says "{default, ...}". So the following are all of the possible commands that you are supposed to run based on your system setup.
Windows & Linux with NVidia Graphics (CUDA and/or TensorRT)
python install.py --onnxruntime cuda
Mac (Apple Silicon and Intel)
python install.py --onnxruntime default
Windows with AMD Graphics (DirectML)
python install.py --onnxruntime directml
Linux with AMD Graphics (ROCm)
python install.py --onnxruntime rocm
Windows & Linux with Intel Graphics (OpenVINO)
python install.py --onnxruntime openvino
i have mac os operating sequia M4 and the python install.py --onnxruntime default does not work. Thoughts?
@TheDustyBlueShow without seeing the error, I couldn't say for sure. Be sure you're in the facefusion directory when running it.
I have a similar problem and he says :python error that he doesn't know the term
Hands down the best software install guide I've ever seen!
@@crippsverse wow, thanks! I appreciate the compliment.
@@shadyendeavor It's the first time I've felt confident about using the terminal for AI, so thanks. Pinokio for my AMD/Nvidia linux PC didn't set up Facefusion that well. So I figured I needed to learn.
@@crippsverse I completely get you on this. I was never a Terminal person. It wasn't until AI became a thing that I "forced" myself to use it. I'd say "learn", but I don't think I have learned it yet. But whenever I had to install something with command line, there was always the worry of missing something and screwing up my computer (which now I know is pretty difficult to do). But now that I've probably done hundreds of installations since and have messed around with a few other things, I've wanted to make sure that whoever did watch my vids, that they had full confidence in what they were doing by laying out exactly what everything was and what it was doing. I hated the videos where people would say, "do this, this and this" and never give you any idea of what each step was actually doing. Anyway, I'm rambling now, but again, thanks for watching and I appreciate the comment.
One step was conveniently left off the installation documentation that isn't always necessary, but may be a problem for some people. During stage 1, once you've done the conda installation, close your terminal or cmd window and open up a new one and continue where you left off.
If this wasn't done, sometimes this can result in following error the error:
CondaError: Run 'conda init' before 'conda activate
Thanks man this was very well explained. Also the pinned updates were useful to me
All are awesome! Thanks for the hard work and the most importantly for sharing with us!
@@alptonay1481 you're very welcome and thanks for watching!
Which face swaping model you suggest
@@ghostatschoolmemories3232 99.9% of the time, the default inswapper is the best - especially when used with pixel boost. I don't really ever do still images, but I think one of them might be slightly better for them.
@@shadyendeavor Well the default (inswapper_128_fp16) work fine but the face detector (yoloface) glitch specially when a person face show half or he or she see above once their eyes are out of sight the detector can't able to keep them up though I really appreciate your work and it's really amazing how far the face fusion has came my suggestion if the tracking can be improve similar with any 3d face tracking
@@ghostatschoolmemories3232 Until better models are introduced, there is nothing that can be done about any shots that are past the profile or not showing enough of the face. It's just a limitation of the current models.
This most likely won't fix these issues, but I always recommend choosing "many" for the detector. This will add an unnoticeable amount of time to a swap, but this uses all of the detectors instead of just one and they all have their strengths.
Can this be used with a webcam for live video?
@@chris20001965 yes, but a powerful nvidia card is needed for it to be worthwhile.
I'm learning so much from your videos. Thanks for time you take in putting together such great content and for the straight forward manner in which you explain things.Quick question. When using FF for lip sync, face enhancement, and frame enhancement for a video, have you noticed that doing each of these three steps separately reduces rendering time in comparison to checking all three boxes and doing them all at once? Thank you again.
FaceFusion 3 is wild..
Hi, everything in the install was going smoothly until I got to the end of Stage 2. After doing conda activate facefusion, I get "CondaError: Run 'conda init' before 'conda activate'". This happened on both 2019 iMac as well as M1 MacStudio. What to do about this?
I'm hoping this is the issue, so try this. But one thing that should be done that was conveniently left out of the documentation is that once you've installed conda, that you should restart your terminal session by either closing the window you were working in or just completely quitting and re-opening terminal and continuing where you left off.
@@shadyendeavor Awesome, thanks! That actually worked! (p.s. love your Brave wisecracks, I totally concur... Brave browser FTW!)
thank you very much, can you do a tutorial on how to update it after maybe from github ?
@@mr.8oom you mean for when say 3.0.1 comes out? Simple, be in your FF directory and conda environment, then:
git pull
python install.py
python facefusion.py run
Prompt for the intro background?
@@8561 the intro background? Do you mean the background that's throughout the entire video of the girl in the tank top with my name on it? I'd have to look it up, but I know it's pretty complicated and uses style references in MidJourney.
@@shadyendeavor Yes that image. I found it hard to prompt for the textured paint blend background style. I can imagine sref helping in midjourney
@@8561 The things I have in the prompt that might help you get what you want are "funk art portrait", "Gil elvgren and jimmy turrell style", and "risograph printed portraiture on artist’s paper". The rest is just adding the text and describing the girl. Well, and the SREF #, which was 3061430968 for this one.
I'm kind of new to this....but I see you used 4 multiple source images, how exactly does that work, and do you have a separate video on it?
I don't have a video that specifically talks about it, though I do have a video where using that feature is a very large part of what the video is about. Check out the following video about morphing in FaceFusion that is all completely done by using multiple source images.
th-cam.com/video/Rz_Zfqn9d2U/w-d-xo.html
Thanks for the video tutorial. I got it up and running. Wondering now why under execution providers CUDA is not showing? I'm only seeing a CPU option right now.
were you not given the option when you did the step of "python install.py"? It SHOULD give you the option to choose from whatever accelerator you installed in step 3.
To try it again, make sure you're in your directory and have your conda environment activate and then install the accelerator:
conda install conda-forge::cuda-runtime=12.4.1 conda-forge::cudnn=9.2.1.18
Then do the install again with that one specifically chosen:
python install.py --onnxruntime cuda
And then do the conda deactivate and activate again:
conda deactivate
conda activate facefusion
If it's still not showing up, you're going to need to ask for more help on the discord. Good luck!
@@shadyendeavor Thanks for taking the time to respond. I realize now the silly mistake I did during installation. Your troubleshoot helped me fix the issue. Cheers!
Did I miss something or didn't you explain WHERE exactly the installation is going to be put on the drive? I collect all my installation in a special sub folder.
You didn't miss anything. I never said where it installs. But with the regular install, it goes directly into your home folder. If you want to have it install and run from somewhere else, you're either going to need to know how to do that yourself or ask for help on the Discord.
Please i don't understand how do you donwload the models?
@@souvenirsdufutur4695 you literally just select whatever model you want inside the app like you're going to use it and it downloads it
great tutorial. thanx.
You're welcome!
hi, thank you for sharing! The upgraded version is amazing! Btw, I have a question about a face leaving or entering the frame, and I wonder if you can advise me on how to avoid "glitches"? Or do you have a video that talks about it?
It's a limitation of the models and there really isn't anything to be done about it other than choosing footage for your targets that doesn't have these situations in it. If the detector models don't see enough of a face, it can't recognize it AS a face. And if it's not detected, the swapping model doesn't know that there is something to swap.
@@shadyendeavor I see. Thank you so much for answering!!!
How do you lipsync a still frame? i.e. include a wav file and a png - then have it animate the still image. Thanks.
You can't with just FaceFusion. The outputs are always based on your target video's length. So if you have a still image, that's all you are going to get. You will have to use a third party software to create an actual video of your still image. Something like Davinci Resolve can easily do this. I'm sure there are other options out there that are much simpler to use if you aren't familiar with editing software.
I've used Microsoft Clipchamp for that exact same purpose. It's a very simple video editor in comparison to the others I have. What I do is put a single image of a face into Clipchamp, or any other video editor for that matter, and just keep on adding the same image repetitively, say for about a 60 second video. I then go back and figure where the 1/3 and 2/3 points are on the total timeline. For the images that are between these two points, I flip those frames horizontally to give some realism to what otherwise would be a very boring video of a single image. I also do a fade transition at the these two points, just to add some flair. Once I have the completed 60 second or so video rendered, I save it to my PC and then go back and put that 60 second video into Clipchamp and basically do all that again to match the video length to my audio track length I want to use for lip synching. It's certainly not the ideal solution, but it does a good job for basic lip synching. Hope that helps.
@@iheartjetaviation2764 since we all have ffmpeg installed for FF, there's a much easier way if you're comfortable with command line. Just use the following command:
ffmpeg -loop 1 -i /Volumes/Prometheus/image2videotest.jpg -vf "pad=iw:ih+mod(ih\,2)" -c:v libx264 -t 10 -pix_fmt yuv420p /Volumes/Prometheus/image2videotest.mp4
Obviously, change the input and output files to the correct directories and names. And then the "-t 10" is for how many seconds you want the video to be. Does it almost instantly.
The weird thing in the middle is modifying the image so that the height is divisible by 2, as you can't create a 264 or 265 if it isn't. Weird, but necessary.
how do I find the output path I want to use and replace the one it shows? video wasn't;t to clear on how to do that unless im just dumb lol
You're not dumb, but also, this video wasn't about the setup or usage of FF. I just put in the quick setup that I used in order to test and show that the installation was successful.
I'm not exactly knowledgeable on the subject, but for me (on a Mac), I have it setup so that it's on an external drive named Prometheus and within a few subdirectories. So my output path looks like this:
/Volumes/Prometheus/Downloads/Videos/FFOutput/
If you simply wanted it in your regular Downloads folder on a Mac (if you haven't changed the location), then it would simply be:
/Users/username/Downloads/
If you're using any directory names with 2 words with a space, that always makes things more complicated. I'm not sure if this works on Windows and Linux, but one thing that is very simple to do is to have a new terminal window open and literally drag the folder/directory of the location you want onto the command line and it should populate with the path. Including if you have multi-named directories.
Hopefully this helps and thanks for watching!
Thanks for this video. Could you please explain what the advantages of using the Windows Installer version of FaceFusion as opposed to using it in Pinokio or via the Chrome link within Pinokio. By the way, I have tried both within Pinokio as well as via the Chrome link (my GPU is the Invidia RTX 3060) and so no reduction in rendering times and the same good quality of the rendered result. Does the windows installer speed up rendering times. Is it worth the money?
For the most part, not really any. Pinokio literally goes through the same exact steps as the separate Windows and Mac installers (and the exact same as this tutorial, in fact). Pinokio has the added benefit of having multiple AI apps all easily available together.
I'm not sure what you mean by the "chrome link". I'm assuming you're talking about the link that will open it up in your default browser instead of running it within Pinokio. If that's the case, then again, no real benefit either way. Other than Pinokio is better as it's not spyware riddled like Chrome and you should be using Brave instead. :P
The installer has no benefit other than hopefully simplifying the install process for those not comfortable with following a tutorial like this one or even comfortable enough to have first installed Pinokio.
@@shadyendeavor I apologize for the delay in thanking you for this explanation. I now can rest easier knowing that I'm not really missing out on anything by sticking to my current method. Love your videos!
@@iheartjetaviation2764 No need to apologize. I'm just happy to help. Thanks for watching my videos!
zsh: command not found: conda
Is it possible to download on Mac 10.12.6?
I could be wrong, but I believe it should be fine. I'm pretty sure Catalina and earlier are the ones that it can't install on because of software dependencies.
For mac how do you locate the .assets/models folder?
To unhide hidden files and directories: command+shift+.
Yes, that's a period at the end. Then you'll see the folder inside the FaceFusion directory.
@@shadyendeavor Thanks
Watching this video I have a bunch of questions. First the execution providers i noticed you deselected cpu and only selected the coreml. I only have cpu, cuda and tensorrt as my options and I atleast make sure cuda is sorking but I make sure to select both CPU and CUDA because I think 2 is better than 1 right? usually when in doing a video face swap i have it running and its taxing my CPU and my GPU both at up to 100% usage. I have Intel Core i7 , 64 GB RAM GPU:NVIDIA RTX 4070 running off a Samsung 980 PRO with heatsink NVME drive connected straight to the motherboard (my root drive) I do have 7 HDD's inside my machine but that root drive is my fastest and big enough at 2 TB's. I also usually push the execution thread count to somehing between 9-11 mostly at 9 and the execution queue count usually at 2 or somethign 3. I also never change the face selector from reference because I never am using anything that has more than one face in it really. ANd I seen that for the source you through in a list of files im assume are all different images of the face you are using, but why do you put multiple fast files in, do you see a difference? And the detector model you picked many as Ive always just picked yoloface just because it seemed like it always was the best as far as speed and quality mix. I also would add occlusion and region to the face mask types and for images i go as far as processing face_swapper, face_enhancer & frame_enhancer. I also pick the 1024x1024 face swapper pixel boost & inswapper_128_fp16 , for face enhancer i think i wouold normally pick comedformer_plus_plus or transformer whatever i know i might be mixing those two up. and ultrasharpx4 for the frame enhancer. but unless its important i dont usually use frame enhacer at all for a video because it slows everything down to nearly a stop from what it seems like.
I dont know anything about what i am doing other than trial and error with every ai tool ive ever started using like auto1111, comfyui, facefusion, rope, forge and even others so im not saying im making the smart choices. All i tend to have are chatgpt plus for help in learning any of it and sometimes watching videos like yours but reading tutorials tend to be almost impossible because i dont tend to retain much information i read and its my slowest form of learning anything. Just wanted to know how wrong people think i am.
First, realize that this was an install video. Not a "how to use facefusion" video. I did a quick setup to see if and show that it was working.
As I say in the video, I'm on a Mac, so I use CoreML and not the other options. As far as between CUDA and TensorRT, I also say in the video as to which one you should choose or go ask for more information on the Discord as I don't know enough about it. As for picking CPU, you only choose that if the other options are not available to you. Picking CPU along with the other options does nothing as FF will only use one. Yes, you will see your CPU being used either way as there are tasks that only run on the CPU. But you choose the accelerator that works best for when there is a process that CAN use it. So no, choosing 2 isn't better than one in this situation.
Otherwise, all of the questions about Face Selectors etc are easily answered by just watching my videos. I have a Basics video that is specific to the Reference selector and another that is specific to the One and Many selector.
Reference - th-cam.com/video/l8D5Of1tLM8/w-d-xo.html
One & Many - th-cam.com/video/e6hMCa4tg1Y/w-d-xo.html
For the detector, Many uses all of the options instead of just one of them. So it has a higher chance of detecting the faces. It might be a bit slower, but nothing noticeable.
I also go over most of the other settings in my Deep Dive video.
Deep DIve - th-cam.com/video/nvPbdHVF12c/w-d-xo.html
1024x1024 is overkill is most all situations. 512x512 is much better to use and takes literally 1/4 of the time to process.
I have a lot of videos on FaceFusion. Best thing would be to just check out my entire playlist on them if you really want to learn as much as you can about FF.
Playlist - th-cam.com/play/PLMwq9QXPN0iDReOut5vKekXT4V6wGTET8.html
@@shadyendeavor oh Yes i know what you are talking about in your video. I just do always pick them both because my amount of time does tend to be a bit shorter with cpu and cuda selected but ill do more experementing. I have been using face fusion for a while too, over a year but like i said just using it trial and error learning. I have been watching alot of your videos the last 2 or 3 days and ill keep it going, thanks for all the help!
Also I was wondering why do you have videos in other languages on this channel and not make a separate channel in each language? It would benefit you more I would think as you can eventually get to monetizing from multiple locations doing that. I just know a few other people that do it that way and it tends to upset people that subscribe and get another language video in their feed.
@@Skettalee if you looked, you'd see that on every dubbed video there is a pinned post stating in that language that all dubbed videos are being moved to a separate channel.
www.youtube.com/@shadyendeavordubs
At this point, there are only 8 videos left on my main channel and I'm moving them over one a day.
thx for the work!
Hello, thanks a lot for the tutorial, can this work on CPU only ? I don't have Nvidia. Thanks !
Yes, it can run on CPU only. But it can also run on AMD and Intel GPUs, The options are all given in the video.
hei guise pls help what os dies windows habe?
CondaError: Run 'conda init' before 'conda activate' I have this error at the end of stage 2 someone can help me I dunno what to do! I tried to activate face fusion but impossible
I use a Mac
@@harisberrouane5688 I got the same error. 😕
Read the pinned comment. I believe that may be the problem.
Possible dumb question but do I have to uninstall the previous FaceFusion 1.6 ? If so, can you make a video on how to do that too ?
Actually, not a dumb question. You have 3 options.
1. (I don't recommend) Just do the install and it SHOULD overwrite everything, as it usually would. However, 3.0 is a big upgrade and a lot of changes were made and even the models need to be changed out.
2. Go to your home directory (which is where facefusion should be installed) and just deleted the facefusion directory/folder and everything inside and then do a clean install.
3. Same as 2, but instead, rename the directory/folder to anything else (even just adding a letter to it will work) and then do a clean install. The new install will ignore the old directory that you changed the name of. Doing this will now keep a backup of your old install, just in case something is wrong with your new install (which there shouldn't be). Once you know everything is working well on your new install, feel free to delete the old one.
Either way, deleting the directory is the same thing as "uninstalling" facefusion. There are no other files to worry about.
@@shadyendeavor thank you so much. I really appreciate the response.
Assisting your algo: I use a Mac normally, I have a hackintosh, about to build a newer one (both with triple boot). I’m investigating a used Epyc system with a plan to use 4 old Nvidia server GPU or 4 newer AMD gpu. ROCm seems to be stable enough for ai gen
I honestly didn't realize that that was a possibility to run rocm on a hackinstosh. That's impressive if you're able to get that to work inline with FaceFusion. Though FaceFusion will not work with multiple GPUs. It can only send data to one. However, you can have multiple instances of FaceFusion running to do multiple jobs at the same time and have each going to a separate GPU. Then I'd just be worried about the CPU being the bottleneck. Or the heat. Let me know how it goes. Would love to know if it works.
@@shadyendeavor I miscommunicated, I wasn’t clear 🙈. The hackintosh is strictly i7 and built in video. I’m not patient enough to tinker with an AMD hackintosh and researching will take time away from using the tool
The Epyc will be Linus and window solely for LLM and stable diffusion
@@peterxyz3541 ah, gotcha, that makes sense. Would have loved that to be a solution for crappy Mac AI. But definitely understand not wanting to take the time to figure that hurdle out. If it's even possible. Either way, good luck with the new system.
it's skipping and not overwriting all target frames when the image is small. So if the target has, let say a mustache, the resulting video will show mustache popping off and on. it also doesn't replace head looking up or down. the face flickers to the previous face.
Need an option to force replace all frames.
Excellent, THANKS !
You're very welcome! Thanks for watching!
sorry to bother you but when i run "python install.py --onnxruntime {default, ...}" i keep getting this error "zsh: parse error near `}'" and now im stuck
you're not putting it in correctly. I even typed it out and put it on the screen 08:50
Whenever you see curly brackets in command line, that means those are the options within to choose from and enter - not to actually duplicate exactly what it shows.
@@shadyendeavor ok thank you sorry for the trouble
@@BobDylan-d2l no trouble or bother at all. Happy to help!
@@shadyendeavor Hi, sorry..I have the same issue. But the result is zsh: command not found: python. I'm on mac. Could you please help me with this error? Thank you
I've fixed it. Just closed the terminal, then run the code again. It works now. Thank you
I got stucked at: Downloading | 0.00/8.00 [00:00
Thank you for this video
I have a problem in step 7, I get this error when excusing the command:
Traceback (most recent call last):
File "/Users/jeremy/facefusion/facefusion.py", line 7, in
from facefusion import core
File "/Users/jeremy/facefusion/facefusion/core.py", line 6, in
import numpy
ModuleNotFoundError: No module named 'numpy'
Thanks for the help
be sure you have your conda environment activated. On the line where you input the run command, does it say "(base)" or "(facefusion)"? If it says base, that means you are not in your conda environment, so type in:
conda activate facefusion
It should then say "(facefusion)" at the start of the line, then try to run.
Otherwise, something else along the way was missed or skipped. But most likely it has something to do with the conda environment.
@@shadyendeavor Yes facefusion is activated, I don't understand I only had an error in step 7 otherwise the rest went well
@@jeremyheyob It still, most likely, has something to do with the conda environment, regardless of it being activated. It's beyond my scope though. I've just seen that most people with this problem on the discord are able to fix it by going through the steps again, which either means a step was missed or input incorrectly, or it means that a step failed somehow.
@@shadyendeavor Ok thank you, I'm going to start again and I'll get back to you to see if it worked, it might help other people.
@@jeremyheyob You can finish install by terminal and reinstall from pinokio, for me it works!
with my poor as gpu, it runs very slowly. Do you have any online version recommendations?
@@taobaofy8500 I'm on an M1 Mac, so most likely, it runs slower for me than you. However, if you want to run it online, RunDiffusion is currently the only option.
Pc requirements?
Pretty much anything above a potato.
But seriously, it's just as long as your system is new enough to actually be able to install the different components. As long as they can install, the app should run. It's just a question of how slowly.
Using simple git pull upgrade, will it work going from 2.6 to 3.0? Also... do I just enter 'default' when choosing my onnxruntime? (sorry... not a coder)
@@QuailRunRoad you'll need to do a new install. Too many of the dependencies have changed. Hence the new install video. 😉
Is there a way without needing stupid conda?
I'm sure it's possible, but you're going to have to ask for help on the Discord for that. I've never had a problem with it, so I'm not sure.
"Hello, on Windows, when I try to run the Conda widget, it doesn't work, and this message appears:
An existing package has already been installed. Attempting to upgrade the installed package...
Failed to upgrade the package using Winget. Use the method provided by the publisher to upgrade this package
Sorry, but that's definitely a question that's out of my league. Especially since I'm on Mac. Ivan or Happyfunball should be able to help you one the FF Discord. Link in the description.
Hey, I just came across your videos after searching for a solution to my problems, your videos are really insightful, thank you for making them 🙏🏻.
I'm going thru a problem , when i face swap, some lines around the swapped face are visible in output video. Can you please tell me how to fix it ? I'm really new to all this.
Thank you, and appreciate you for your time.
If you're referring to when it kind of looks like a ghost frame around the sides of the face, I get that a lot too. Try switching between using "occlusion" and "box", or just one of them if you have both turned on. Hopefully that should fix it. Otherwise, depending on the footage you're using, it might be possible to use the masking to cut it off. You'll have to turn on the debugger in order to see the masks and then try to "shave" off the ghosting. Pain in the butt, but not much else to really try. Just make sure you turn off the debugger when you render out the video/image.
@shadyendeavor Hi, thanks a lot brother, changing face mask type to occulsion helped. It worked perfectly. Thanks a lot again, really appreciate it. 🙌🏻
@@sharad_10 You're very welcome. Have fun!
Has anyone faced a situation where the original video is a 16-second video and the face swap only generates a 2-second video?
Its says Facefusion - Error in installing cloudflare on mac. Why?:) Pls help
I'm sorry, I have no idea. Best to try asking on the Discord that's linked in the description.
All seemed to go ok, but upon running, I see no execution provider other than my CPU. It swaps images fine but videos aren't working (presumably because it takes forever).
My thinking is that I needed to use the --onnxruntime cuda command, instead of default. Now I don't know what to do. I'd like to uninstall and start from scratch but then, I don't know how to uninstall either. I'm going to risk just renaming the facefusion folder and doing another from the start.
@@GlynnC-q2k read the pinned comment 😉
@@GlynnC-q2k uninstalling is just literally deleting the FF folder. But you don't need to, just do the step in the pinned comment.
Pleased to say, that all worked. Rename the original folder, reinstalled from scratch and used the 'cuda' option in the install instead of 'default'. All working now :)
@@shadyendeavor Thanks for replies. Sadly, upon closing and opening again, it again doesn't have cuda option for the accelerator :( It did after the 2nd install, but now it's gone again.
I INSTALL IT ON WINDOWS 11 ON LINUX MINT AND I EVEN BOUGHT THE WINDOWS INSTALLER FOR 20 BUCKS NO VERSION IS WORKING. what iam doing wrong
I have no idea. There are way too many factors that could be the problem. Your best option is to contact the actual developers on their discord channel which is linked in the description. And please, stop using caps.
Is it normal that a 13 seconds video requires something like 8 hours??? (output 1280x720 quality 80)
There are so many reasons why that could be "normal" and not. It depends on many different settings. First, what kind of computer are you on, Mac or PC? Are you set to only CPU? Did you install the accelerator correctly as stated in the pinned comment? Are you only set to Face Swapper? Do you have Face Enhancer turned on? Do you have Frame Enhancer or any other models turned on? Do you have Pixel Boost turned up?
🙀..no Premiere...???? this is outrageous 😾 [Kitty launches into tirade begins clawing all furniture in home 😾]
.
FaceFusion 3.0 Epic Trailor 👉 th-cam.com/video/2QzTj6o-pv4/w-d-xo.html
@@NinjaKittyBonks sorry you didn't get to stretch your moderation claws.
@@shadyendeavor …. That’s GODerator, lest you want rhat perma-ban I warned you about 😾
you are my god
Wow! Thanks! You're giving me quite the responsibility. 😛
DONT WORK ANYMORE. SORRY DUDE. BUT THX.
@@padawanbennekinDE it absolutely does work.
@@shadyendeavor What are the dependencies??? It run and prozess but show no output. On win 11 I dont get it to run. It is apain I tried so much cost me days.
@@padawanbennekinDE what does your terminal say?
@@shadyendeavor I switch now to linux mint . Here the same no cuda button running step one 100% then nothing.
Unfortunately this is censored
How can you censor a face?
where is the face swappers for videos?
@@AI-Consultant I'm not sure what you're asking. All face swapping models in FaceFusion work for videos and still images.