วีดีโอ

You're right. Original checkpoints

Sorry. There is an error in readme.txt. - 3. CD to your simswap cloned directory and paste the commandline.

@@heinzelmann1963 When I open the SimSwap, it says this and doesn't open 👇👇 'component 'wmp.dll or one of its dependencies not correctly registered:a file is missing or invalid

After cd simswap how to run GUI

@@__fifaworldcup8857 This is an VB6 on Win10/11 problem. Copy 'comdlg32.ocx' from package folder 'syswow64' to the corresponding windows folder, then try importing 'vbcont60.reg' to the registry. 'vbcont60.zip' contains some more VB6 system files if neccessary.

@@heinzelmann1963 Thanks bro ❤️

No. It's a local installation running a rewritten inference. It detects, crops and aligns the face frame by frame for animating the image. Animation then replaces the original face using face parsing mask.

HI. How do you change the face of a full figure? Do you use two images American camp and Ted option?

This is only for local installation.

There is also a script for face swap from image to image on simswap github.

I never tried deepfacelab so I can say nothing about it

@@heinzelmann1963 okey thanks!

Yes

@@heinzelmann1963 Thanks! Is there a way to use VGGFace2-HQ with this? Also, curious do either of these GUIS work better than Kinter? Seems functionality is the same no?

@@Because_Reasons It is only usuable for simswap. For me it's easier to do such things in Visual Basic. I have another simple simswap version written using TKinter... You can find the download link in the video description

Maybe I'll publish the script and all additional files (face-detection, face-parsing, face-alingnment...) when I have cleaned the code and written a short manual. Basically it can be run like the original but it has a lot of parameters for finetuning...

@@heinzelmann1963 Excellent, I will wait for your manual. I recently started to get interested in neural networks and video face replacement. I would be interested in working with your code

You need an already working original SimSwap commandline version. You can not run the example batch file without editing the paths. Give me more details how you have tried to run it

@@heinzelmann1963 the same thing happens in my original SimSwap, it just goes to end and does not continue, I've seen other people have this problem but can't find a solution.

Sorry, I think I can't help you with this problem. I'm not involved with the SimSwap project on github. Do you mean the "stuck ad end" problem? I can't find it in issues on github. Does it show any error message?

@@heinzelmann1963 If you can't help its ok, but there is no error message, the command line just goes to "End", like it is supposed to, but does not go from there

So it's issue #186 ? github.com/neuralchen/SimSwap/issues/186 Sorry, can't help

Yes. SimSwap needs an Nvidia graphics card

Do you mean increase the bbox around the detected face or increasing the final faceparsing mask? To increase the mask you can try the cv2.dilate function at the end of return2original.py But that has no effect on the result

@@heinzelmann1963 hi yes - thanks for your reply. It was more about how to increase the mask so more of the face is transferred rather than just the main features.

If I remember correctly you can also do this in 'encode segmentation' or 'postprocess' function in the script reverse2original. You can also play around with the face parts ID that you want to swap.

@@heinzelmann1963 thanks Thomas - I will take a look :)

@@farazaslam7445 did you find a solution for expanding the bbox? Need it for a similar project.

Hi Thomas de can you please guide me how to crop face in video And rejoin again in the same place in full frame of video.. I download cropjoin.py but I don't know how to add these file in colab and how to run and also guide me how to use facial recognition to find faces and add multifaces in video through this method .. thanks in advance 🙏

@@adeelnazar88 The cropjoin.py is just a simple example how it can be done. Reading frame by frame, detecting, cropping, do something with the cropped face and overlay it to the full frame again. So theres a first step (crop) and a second step (join). You have to insert them into your specific code... Search github for similar things. It needs some kind of learning how to... sorry

Thomas de can you share me your new modified co segmentation version which does all thing automatically

@@adeelnazar88 Maybe I make a download package with all files needed. But you need a working anaconda environment of the original GitHub repository to use it. And it got a lot of additional commandline parameters, but most of got a default value to start with

They do actually. Boy Harsher and Nuovo Testamento are two examples that come to mind

The disabled buttons write a batchfile to run SimSwap directly from the GUI. But it depends on installation paths: - Run anaconda - Activate SimSwap environment - CD to SimSwap directory - Run the SimSwap python script..

Even Wav2Lip-Emotion with the CROP Thomas use on his GUI will be already much simple to use for sure! I hope he will consider it since Wav2Lip-Emotion is impossible to understand how install and run using Anaconda and Windows at the moment. Great suggestion btw!

@@MrDanINSANE True, the original Wav2Lip was simple and easy to install but their pre-trained model is very low quality, and it's not dynamic as Wav2Lip-Emotion which is INSANE! but as you know not possible to install on Windows / Anaconda because there are no tutorials or step-by-step documentations. GUI will be great improvement but still many people don't understand how to actually install it on Windows...

I've installed the original Wav2Lip about 2 years ago and of course made a GUI for it ;-) Got it to work on longer files even with only 2GB GPU... Biggest problem seems to be mouth movment if there is silence in the wave file.

@@heinzelmann1963 Sounds really cool! GUI in general for Anaconda / Windows from github, especially for video apps. For the original Wav2Lip I created series of batch files since I'm not a programmer, but it was really much easier to run while all I need is to put the files I want in 1 directory, and run the command. True, when there is silence I thought they should've make a "close mouth" but I have no idea why they didn't, it makes more sense. For FAST movement for example, it's recommended to use "--NoSmooth" command or something like that (I don't remember) so it will be more accurate even on BIG movements between frames, I think YOUR crop tool will be a bigger improvement since it's visual. Another problem the original "Wav2Lip" have is the angle, if you have a face with more than 25 degrees rotate, it will either not recognize it or it will.. but your result will be a 90 degrees mouth which looks WRONG hehe :D For the amazing "Wav2Lip-Emotion" I couldn't even understand how to run it in general, their instructions are... not very clear and their not updating it or explaining how to install and use it. So if you'll be able to have a look on it and do some magic GUI for that version it will be very cool, I hope you'll also make a video on your channel to show "how to" + some results, since nobody show anything about it for a long time...

@@anokmaran2845 maybe this could be worth a try. th-cam.com/video/6J_28OKeqLk/w-d-xo.html

Grundsätzlich ja, falls ich es schaffe das Projekt zu einem Paket zu bündeln bevor Putin vollkommen durchdreht...

@@heinzelmann1963 Thanks. I was looking for python script which tracks and crop the head and ability to paste back after some processing(swap for example). I think you are working for windows package right? I'm on linux. We can hope war will be over soon.

@@bluefoxcreation yes, I'm on windows but it should work for linux too? That's exactly where I started from. Track - crop- do something - join. If I find one of my first simple scripts just doing that I can send it to you or post a download link. If you are a bit familiar with python you should get it to work

@@heinzelmann1963 yeah that will be helpful

@@bluefoxcreation Ok. This is the initial basic script for facetrack crop/join. Uses caffemodel for tracking, no face masking or rotation. All needed files included. Usage "python crop_join.py --video video.mp4 --output output.mp4" - Needs cv2, argparse and numpy. Hope it will help you. Link: drive.google.com/file/d/1AhB40ZM89UMVyeR3_fIUBGI79xT5BfT1/view?usp=sharing

I will add drag'n'drop next update

@@heinzelmann1963 That will be great! Any chance to also add in the future the other options such as: "Multi-Specific-Faces" so we'll be able to select more than 1 face at the same time based on extra images? For example Image 1, Image 2, Image 3 will work based on the order of how you crop in your GUI face selection (which is GREAT! btw) Keep up the good work!

Added drag'n'drop. Hope it works

@@heinzelmann1963 Thanks for the update, You Rocks! :) I just tried it and it seems to work, I must say it's SO MUCH EASIER to use now thanks to your GUI! Any chance that the Drag & Drop area will be on the BIG block instead of the grey area with the text? (or both?) just an idea. I hope you'll consider to check out Wav2Lip-Emotion for the same process you did which supports Windows + Anaconda, that will be great! (Wav2Lip original is nice but not powerful as the Emotion one). Keep up the good work!

@@heinzelmann1963 Thanks!

Yes, you are right. SimSwap can do a good job in many situations. But like other methods of face swapping it depends on the used material and what you want to achieve. SimSwap does not swap the face image itself but the facial features, it's more like a warping the original face. Other attempts, triangulation of the image using Dlib (68 points) or Mediapipe (486 points) deform the face image and merge it into the target video. That causes other problems. And none of these methods can swap the hair, very important thing to recognize a person I think. Most realistic can be using Part Swap, but only if there is not too much motion. I'm not involved in the SimSwap project, I only made some GUIs for different face swap projects I've found on Github, so I can't tell if there will come improvements.

I guess you've downloaded the SimSwapGUI3. The download contains a folder called SimSwap. You have to copy all files inside this folder to your local SimSwap root directory with the same file structure. This will replace some of the original files ! ! There are two versions of SimSwap and the GUI ! You can NOT use the SimSwapGUI3 with the latest SimSwap version (512 HQ update november 2021) and vice versa. If you cloned the 512 HQ version of SimSwap you have to use the SimSwapGUI512

@@heinzelmann1963 thanks for clarifying.. I think that's what I did.

I must admit that it's not really calculated.... it's more an experimental value as starting point (if the head is filling the frame height its just 1/3 of the frame height) Thats why you can edit that parameter. I haven't looked at simswap code for longer time because I'm currently optimizing my modified version of part swap / motion co-segmentation

If you want to enter the kernel_h manually when the script started, you can add print ("kernel h : " + str(opt.kernel_h)) opt.kernel_h = int(input("Enter new value: ")or opt.kernel_h) in video_swap_single_instant.py before 'video_swap(...... Generally the tracked/cropped faces (not the full head) internally are resized to 224x224. So a value of around 70 should do the thing

@@heinzelmann1963 Thanks for the answer! I meant how to get and show this starting point from which you can start entering the value of kernel_h as in your program only using python. This is a really useful option and would save time if you could see it when you run the script and starting from it, select your more suitable values.

@@heinzelmann1963 hey Thomas, had a question for you. Why did you choose to change the erosion kernel size instead of removing the face id parts? For example index # 6 being the nose I believe. Also the kernal height are you referring to the det size in test_videoswapsingle? Which value would you change on the python script to shorten the erosion mask? Thanks buddy, great work btw.

@@Gh0st_0723 removing face id parts sometimes does not work very good. Det size has nothing to do with it.

You have to manually register comdlg32. On some win10 systems this seems to be not possible. Maybe the vb6 runtime files on this page can help you. The 'vbcont60.zip' also contains a registry file for comdlg32. Don't know if it works...maybe you give it a try (at your own risk) schierlm.users.sourceforge.net/mschierlm/smsoft/runtime.htm

@@heinzelmann1963 I have rtx 3050 when i try to start script it stucks at end do you know the solution ?

@@Maeve472 Sorry, can't help. Visit simswap on github / issues. Many others seem to have similar problems with RTX series

@@heinzelmann1963 I can't get past this issue. Is the GUI incompatible with windows 10 onwards? I have tried my best to install that ocx but I get a message "this module is not compatible with the version of windows you're running".

@@beyshadow vb6 is outdated an can be tricky to run on win10. Try the latest download. I added some vb6 setup files. Try to import the .reg file. Maybe make a backup before?

For preview in half resolution of the original video add the following lines at the end of "reverse2original.py" (h, w) = final_img.shape[:2] h = h//2 w = w//2 final_img = cv2.resize(final_img, (w, h)) cv2.imshow("Final Video", final_img) cv2.waitKey(1) Same indent, use space not tab! You can use notepad for that

@@heinzelmann1963 Thank you!

Luxand Faceswapper does a good job. And some Github repos...

Thank you for your reply. Unfortunately I can't help you with colab. (Doesn't work for me) This version is just intended for local installation, but you can compare the original files to the ones I've modified. There are quite some more differences. I'm not involved in the original simswap repo

@@heinzelmann1963 I trying to add your files into collab and edit SimSwap colab.ipynb but have an error. Same when i compare and edit original files. Perhaps you can tell me in which original file I can manually edit values ​​like --kernel_h and --kernel_w? Whatever, thank for reply.

@@netrunner-exe in "reverse2original.py" line 137 you can change the values kernel = np.ones((40,40),np.uint8) (40,40) are (kernel_h, kernel_w) given from the commandline in my modified script. Only change the numbers, eg. (80,40)

​@@heinzelmann1963 It works, thank you!)