Deepfake Tutorial and Explanation Step by Step GPU/CPU
ฝัง
- เผยแพร่เมื่อ 15 ก.ค. 2024
- In this video we will look at how to create a Deepfake using Deep Face Lab. Using a simple cheat sheet we will go step up a step to train our model to create a realistic-looking Deepfake.
🚀🚀 My Urdu/Hindi AI TH-cam Channel 🚀🚀
/ @murtazahassan01
Course and files:
coming soon ...
Deep Face Lab:
github.com/iperov/DeepFaceLab
Premium Courses:
✔️ Computer Vision Game Development Course:
bit.ly/3ttLZ2s
✔️ Computer Vision with Arduino Course:
bit.ly/3wzLB4m
✔️ Advanced Drone Programming Course:
bit.ly/3qs3v5g
✔️ Learn to Build Computer Vision Mobile Apps:
bit.ly/3uioY1J
✔️ Jetson Nano Premium Course:
bit.ly/3L8uIlF
⚙️⚙️⚙️--My Gear - ⚙️⚙️⚙️
👉 Complete Gear 💈: www.computervision.zone/tech-...
👉 My PC Specs 🖥️: www.computervision.zone/tech-...
👉 My Video Shooting Gear📽️ : www.computervision.zone/tech-...
👉 My Laptops 💻: www.computervision.zone/tech-...
👉 Educational Products🧑🎓: www.computervision.zone/tech-...
👉 TH-cam Starter Kit 🔴: www.computervision.zone/tech-...
Follow Me:
TikTok: bit.ly/3Vo76OQ
Facebook Group: bit.ly/3irDcb7
Discord: bit.ly/3JvyxAM
Facebook Page: bit.ly/3IvpU7W
Instagram : bit.ly/3NdGME3
Website: bit.ly/3ICFTS0
Github: bit.ly/3woU6PS
#ComputerVision
#OpenCV
#CVZone
00:00 Introduction
02:55 Files and Downlaods
09:34 Theory
16:40 Step 0
17:45 Step 1
18:04 Step 2
19:47 Step 3
21:17 Step 4
23:49 Step 5
26:50 Step 6
32:50 Step 7
37:28 Step 8
42:10 Result
Disclaimer: This video is for educational purposes only. Any action you take upon the information you find on this tutorial is strictly at your own risk. This channel will not be liable for any losses and/or damages in connection with the use of this tutorial.
Boss , serious u r taking teaching to a different level . Cant believe we r getting learning like this for free
Glad to hear that.
Is this course is available for free!?😱😱😱
@@athiamaanr Did you pay to view this video? /s
@@tunesoop7653 well, he actually paid with his "view" and Data. :-P
@@murtazasworkshop at 33:20, i leave blank like you say for use interactive merger i press enter, number of workers i leave blank press enter, it then says collecting alignments and computing motion vectors, its hits 100% but then the grey merger box with all of the buttons to manipulate does not pop up anywhere for me, do you know what may be causing this? thank you
Thanks sooooo much for you tutorials! Just started understanding this field of programming and your tutorials are making it so much easier!!!
This is epic. For a person to learn all of this themself, it would take anywhere from a couple of weeks to months. You distilled all the information in just a 45 min video! You are really a generous genius. Hats off to you.
Can someone tell I was training saehd laptop turned off cuz No charge now will it be saved or not
Very nice tutorial - lots of ins 'n' outs to deep face labs and your tutorial was one of the more helpful ones, thanks!
Brother,
Everyday i wake up and check the notifications, that have you uploaded any videos !!😍❤️
Thanks so much for this in depth tutorial. I was finally able to complete the example, and I am ready to try my own!
Excellent sir.. I am following and practicing all your teachings. Wonderful collections sir
Great Work Bhai...
Keep it Up..
Day by Day You are becoming an Inspiration forme to Join the AI field.....
God Bless You.
Love from 🇵🇰
Amazing Mr. Murtaza. Perfecto
This is a very straight forward and detailed video, very well done and straight to the point. Thank you for the help!
Very clear and understandable. Thank you
Wow am super pumped to learn this
Love the videos interestingly cool. Love the sense of humor too.
This is super helpful and you are so generous to teach us free!! Cudo's to you!!❤
Thank you very much for this incredible tutorial
Great tutorial..has been for it very long
Crazy level content. AND Free. Great Job Man 🙌
depply waiting for this tutorial thanks man..............
Great man, great video tutorials. Thanks a lot sir
Incredible great video. Thank you for sharing.
Your accent sounds very pleasant and clear, didn't expect that ;)
Thanku sooo much for this tutorial .....this is soo amazing ...
thank you so much! Your tutorial is so perfect.
Well explained. The best I have seen so for on this topic. Thanks so much.
Already the second time I'm watching this and it's so amazing! One question, what would happen if I choose "head" in the wf Settings? Can you do a tutorial on that, since this function seems pretty underrated. Thank you so much. :)
My dear sir, you are a savior. Thanks for this tutorial.
This is gold! Many thanks from Nairobi, Kenya
Thank u very muck. Very good explain algorithms. You are a greate man
Muy buen vídeo, cubre los aspectos básicos para que quede muy decente!
Awasome as always
Thanks for this!!
Best tutorial about DFL, I learned about the new functions, I did not know that I need to press Shift + / , to apply the config to all other frames.
You are my best teacher.
Thank you sir for such a great tutorial love from Delhi
Super video. Thanks
Thanks! This video was very helpful. There was a bit of confusion with some things like the "up arrow" on the keyboard in the diagram but it is actually the right hand shift key that has the little up arrow on the American English keyboard.
When he say's to switch screens during the merging process, it sounds like he is saying "press on top", he is actually say "press on tab". And the Tab key typically just says Tab in English and we never look at the little arrows on it. So I was thrown off. Haha! I'll share more little quirks as I go through it to help other English speaking Americans.
A little tip. The software doesn't work when I use #7 which is "merge" when I select GPU. I have had to use CPU. I also had to update my video card driver to get it to actually use the GPU during the initial process where it scans the faces. It was worth it to get the GPU to function over using the CPU. It is about 20 times quicker. I let it run on GPU for 20 hours and it had 28,000 iterations. I am getting about 5,000 iterations per hour with my Nvidia GTX 1660 and an Intel I 4770.
@@markfothebeast great advices! Thank you for sharing!
Congratulations Bro!!
Great video. Thanks
Sir amazing this is I can't believe
This is the best Tut ever. Even the original source don't have this kind 0f tut
I checked - everything is clean
Nicee video's lovee itttt gonno use this for my final year project
Good luck
Bruh thank you very much... But make disclaimer that no one will use this for doing wrong of someone
Thanks will do.
@@murtazasworkshop Yeah
Please make a tutorial on Lip Sync from Audio. It will very helpful to us. BTW great video. It helps me a lot.
Great!
Very very cool...
You are awesome sir.
You are awesome brother.
Hey! Thanks for the tutorial! One thing I’m wondering is can you pause the encoding and generating the faces? Let’s say if I have to shut down the computer. Only with quick96?
Really good content! Can you make a tutorial for Aliaksandr Siarohin's First Order Motion Model?
I would love to see an explanation of the different training algorithms
Hi, does the source video lighting of the persons face have to be similar to the merging video lighting for best results or would a brightly lit source face work just as good?
So nice👏👏
Hello Murtaza thanks for the nice understandable explanation, I would like to ask you I am working on a short documentary about my grandfather who passed away a long time ago is there any chance to use static photography of his face and put it on the face of the actor (me)? Or it is some possibilities to pick up a different angle of his face from several old photographs and make a image sequence out of them or video sequence and use it? Thank you.
Zabaedest😘
Thank you
HATS OFF!
Thank You So Much My Brother😪🥺🥺
not sure if somebody mentioned this before... But the elephant Lion metaphor to understand encoder/decoder relationship needs one more step to make sense.
After the student learns to draw an elephant,
missing step -- > " we teach him how to draw a Lion"
Now, when we draw an elephant in a certain pose, the student should be able to draw a lion in a matching pose.
Great video overall. Thanks a lot and keep it up!
great
wow u are awesome and you should give your website link in the description also bro Thanks :)
This is the best video I've watched so far on the topic. Learning this process, I have a question. How can a mobile app takes just a single photo and be able to achieve a really good deepfake using just a few seconds? I know all the cloud processing stuff, but there must be something else, right?
I thought about this as well!
Do a video explanation of xseg markings and training
Great
Great bro!!! Can you work on any deepfake detection one ?
Thanks for this learning video. Can I use available inages directly instead of extracting them from a video file?
Thanks a lot!! Does the reface app use the same model??
Respect
thanks
Please make some tutorial for the usage of aruco markers 🙏🏼🙏🏼
awesome
how are you dear do you have email address to reach up to you or a whatsapp number
awesome :-)
Thanks
Hey bro, I just wanted to ask, are you going to make a intermediate tutorial about deepfake?
Like for example: How the manual option works
anyways new subscriber here hope you have a pleasant morrow!
My thanks to thee.
I assume it's best to use videos with matching frame rates? Or does it not really matter as long as your destination frame rate is your preferred frame rate?
Thank you so much for making our life easy you explained it so well, I am following your tutorial and it is working fine but in my case, at step 4 where you extract the file (data_src faceset extract.bat) the GPU/CPU selection option is not appearing, it starting from 'Face Type' selection. If I am not wrong it is not utilizing my GPU but how can I fix it? any help will be appreciated. Thanks in Advance.
Do all the source images have to come from the same video or could we merge a few videos for a better variety of angles and lightiong to get a more realistic output?
LEGGEND THAAANKS
how are you dear do you have email address to reach up to you or a whatsapp number?
at 33:20, i leave blank like you say for use interactive merger i press enter, number of workers i leave blank press enter, it then says collecting alignments and computing motion vectors, its hits 100% but then the grey merger box with all of the buttons to manipulate does not pop up anywhere for me, do you know what may be causing this? thank you
For anyone else having this issue - increase your pagefile to 4x your RAM i.e. for 32Gb of RAM, make your pagefile 128,000Mb. This fixed the issue for me.
pretraining in saehd, you know where your model files are, you can keep them for later, people share pretrained models etc. what does pretraining in xseg do? what's its difference from saehd's pretraining? where are the model files for xseg pretrained model? and how come people don't seem to exchange these models like they do saehd models? and finally, saehd trains the source model. does xseg also pretrain just the source model?
Thank you. What about cfg to the previous frames?
murtaza is a life saver
merci bg
Nice work man. thanks you so much!
have a question. because of training of models. How can y nake it autimatically? Cause after hour of holding a button P its only 3500 iter points.
also, do you know why smetimes on merge part program just stucking and not responding aymore
and last question. what to do if i have different rakurs and face that i wanted to change really small. In that case program do not working and face stay original.
will be really glad if you will have time to answere. Thanks for greate work. Peace!
awesome
))murask
Could you make a tutorial on how to use a pre-train model
Hello , Thanks for sharing such a detailed tutorial , Murtaza i want my cartoon character speak with deep fake but it replaces cartoon face with my face , can you let me know how is it possible ?
Someone wrote me to reverse it but honestly i could not understand
Nice tutorial. Is quick96 faster or better than saed? I've not had great success so far. I downloaded some footage off the Web, did all the procedure except with Saed. I let it run over night and had 200k iterations but it was still quite poor.
Also, I just heard recently that running deep face lab burns out hdd quite fast due to all the extra use. Which is quite worrying. One person commented they had got through 2 samsung pro ssd's in a year!
That linus one though
Is it possible to change the words of the speaker or does this only work for the original audio?
I've uncovered a bug and a work around if anyone else is having the same issue with the current build as of 2/1/21.
For some reason, I need to keep pressing S or P in the training window every few minutes. If I don't the program will stop training, and become unresponsive to S, P, and Enter commands (however the space bar still works). I don't know for sure but it appears to be any command sent to the Training Preview that gets passed on to the command prompt. If that happens, all the new training information is lost because I can't save it and need to force quite.
The work around I've come up with is using Pulover's macro creator (a free automation app) to just send P (wait a few seconds) S to the training preview window every 1-3 minutes so that I can go AFK and let it run while I sleep, eat dinner, work out or whatever. It's annoying, but I can't complain too much since the software is free and producing good results.
Sir plz tell how to learn basics of Computer visions. Plz sir 🙏🙏
Im wondering what happens if you were to put multiple videos in the scr folder? Say (sylvester stalone) rocky 1 2 3 rambo etc. I know you hav to block out other faces...
Also what about images? Dropping a load of high res google images of sly stallone face?
Where's the software that have an UI that you can select source data and destination data and then click run and then you get what you want ? DeepFaceLab looks like a rocket engineering, man ! I know it could be learned, but it should be much easier ! After Effects also can do the face replacemant, if the user knows how to operate After Effects.
Hey bro,thanks for the tutorial firstly…brill job,but i cant seem to get past step 2,when i extract images from vid data src,i get lots of errors saying failed to load NATIVE TENSORFLOW RUNTIME??
can I use the created model for other videos swap as well??
I men same faces but different videos, can we do it?
Brother Well Explained ! Can we Use deepfakelab on Amd Rx 560 4gb GPUs. or Amd Graphics card
Hello, I had the opportunity to work with Deepface Lab on my PC icore7 4770 Haswell 3.40ghz + 16RAM, by default with CPU it is very slow, but I find it annoying that to extract faces from a 7 minute video I have to wait 136 hours and in the Quick96 training / model exceeding 150k interactions very few parts of the video look "considerably good", the rest look fake and blurry.
Soon I will buy an "NVIDIA Geforce GTX 1070TI 8GB ASUS GDDR5" GPU, will Deepface Lab be able to work with that type of GPU, will I be able to process everything faster and work with XSEG training in SAEHD?
With quick96 it is not what I expected, although I had 29,000 faces in data dst and in data scr 80,000 extracted faces, with different types of lighting, quality in 720 and 1080p, some in 4k.
im stuck at step no.5, i can extract face from src but cant do that from dst, im using cpu, i have intel uhd-none of them work,sys error msg: "cpu doesnt respond,terminating it"..can you help me?
Hi can you please make a video on liveness (spoof) detection
Replay and image attack
Help me please. How to make the object (microphone) higher than the face?