MrDeepFakes Forums

Some content may not be available to Guests. Consider registering an account to enjoy unrestricted access to guides, support and tools

  • We are looking for community members who are intested in helping out. See our HELP WANTED post.

couple of questions from a returner

d33p3r

DF Vagrant
Hi guys, I used DFL last year and did not care anymore because results were not as expected. I am impressed what have been done so far and the DF topic catched my mind again when  sawing the home stallone fake on youtube.

I have to say: I want to make a deepfake with model src and pornstar dst and body simmilarity is most important so im limited in choosing the right face structre. but my dst pornstar is pretty close, the upper half of the face is similar to dst, the lower half is bigger, bigger chin, bigger mouth, to mention that

now I went by the old guide in the DFL 1.0 outdated thread 'My Personal DeepFake Workflow Using DeepFaceLab' from dpfks. I already went hard on sorting just the best src images, min res. 720p and only sharp images without shadows etc I think I did a pretty good job on this one.

the dst was harder for me because I have angles on the pornstar where camera is looking on the head pointing on the forehead or very from bottm pointing at the (double)chin. landmarks on these were faulty so I re extracetd manually but even here some of those are simply not correct landmarked no matter how I scroll or move the mouse. I tried the nearest and saved it.

I need to mention: I have very much angles of the src model from interview clips and such, but getting those very up and bottom angles is very hard to find so actually I dont have those very unusual angles in src. first question here is: If I find some how can I add them to my src and continue training without restart the whole model? 

Now for settings atm I am using:


Code:
================= Model Summary =================
==                                             ==
==            Model name: model2porn_SAEHD ==
==                                             ==
==     Current iteration: 123507               ==
==                                             ==
==--------------- Model Options ---------------==
==                                             ==
==            resolution: 128                  ==
==             face_type: f                    ==
==     models_opt_on_gpu: True                 ==
==                 archi: dfhd                 ==
==               ae_dims: 512                  ==
==                e_dims: 64                   ==
==                d_dims: 48                   ==
==           d_mask_dims: 16                   ==
==            learn_mask: False                ==
==            lr_dropout: False                ==
==           random_warp: True                 ==
==             gan_power: 1.0                  ==
==       true_face_power: 0.01                 ==
==      face_style_power: 0.0                  ==
==        bg_style_power: 0.0                  ==
==               ct_mode: none                 ==
==              clipgrad: True                 ==
==              pretrain: False                ==
==            autobackup: False                ==
== write_preview_history: True                 ==
==           target_iter: 0                    ==
==           random_flip: False                ==
==            batch_size: 12                   ==
==                                             ==
==---------------- Running On -----------------==
==                                             ==
==          Device index: 0                    ==
==                  Name: GeForce GTX 1080 Ti  ==
==                  VRAM: 11.00GB              ==
==                                             ==
=================================================


at start I had batch size 8, unfortunately no clipgrad enabled, gan power 0, trueface 0, face style 10 bg style 10. changed at about 80k its like mentioned in the old HowTo. Now I have very good preview results for windows 2 and 4 where DFL just build the given face. I also have some nice results in the  last preview windows but most of them are still bad. sometimes the chin of the result is the other direction than the dst face, sometims its just a whirled messed up pixelation. in those cases i think it is because I dont have the same angle in my src.

i also have the problem that the result is not looking in the same directions as the dst face (eyes). skin color from src is taken but I didnt experiment with color options yet. Some dst faces have head down and some hair in front of the face, landmarks are ok, should I deleted those?

I have like 163k its now with his model and I wonder if I can add new src faces with missing angles to fix some problems? I also thinking about if I should just delete those dst faces with misaligned landmarks that I can not correct? I wonder if I do this if I have to retrain?

I also wonder if LIAF would be the better choice than DF for my project or if I messed it up by using some options in the start and/or left out later? in terms of loss i am at [0.3038][0.1437] should i wait train more or will less loss never fix angle/eye problems?

regards

edit: attached some examples. Here you see dst looking down getting filmed from above:

https://ibb.co/DwMn313

tried to fix chin area which has bad landmarks obv but wasnt able to

that obv results in a mess:

https://ibb.co/Hxp00Wj

another one where chin is other direction messing up everything:

https://ibb.co/2dDWSgX

on these problem I thought it is just slightly from down angle and I dont have that in src faces but still I have much faces looking in the left direction so it should make it?!

side profiles are very good btw. dst has bigger nose cheeks and chin but DFL manages that to be src shape just perfect.

edit2: one more question: when manually setting the landscapes should the blue bottom edge end one the real chin or the double chin when dst shows a double chin? do I have to retrain from scratch or can i redo dst files and train mistrained model?
 

TMBDF

Moderator | Deepfake Creator | Guide maintainer
Staff member
Moderator
Verified Video Creator
The blue box and it's size is not relevant, it's just to indicate roughly which area it is looking at to find face but sometimes you need to scale it up or down quite a bit or move in a weird place for it to correctly detect it, red box has an arrow that indicates which way is "up" and that should obviously point to forehead/up and the box itself should contain whole face, you don't want it too zoomed in or out. Landmarks should always land on the edge of the face.
 

TMBDF

Moderator | Deepfake Creator | Guide maintainer
Staff member
Moderator
Verified Video Creator
No, just rename aligned in data_src to something else and put new faces in data_src (or extract new from new data_src.mp4) and then run the extractor, at the end cleanup new aligned folder and then copy over or merge both folders (select rename when copying over from one folder to another so you don't overwrite any).

More useful tips in FAQ: https://mrdeepfakes.com/forums/thread-frequently-asked-questions-dfl-1-0-outdated
 

d33p3r

DF Vagrant
so I added some faces from extreme above angles to source and placed them into data_src/aligned. is it necessary/essential to have also the right names in the debug folder similar to those in aligned? I guess because I sorted somewhere before they do not match anymore.

Next again I deleted bad landscapes in data_dst/aligned_debug and manually re extracted these. I found pressing a button helps a lot to get the real extreme angles captchured correct this time. made a backup of my fst/aligned before, then added the new extracted faces to my backup. it overwrote some old files but again not all, I sorted out before I guess. I guess either now I have some bad ones that did not overwrote now.

hard to say, might redo dst completely. I will continue to train now for some more time lets see if it helps. settings as mentioned above since I might have more src than dst faces and pretty much angles in src I dont think random_flip will help and I will keep random warp enabled because I still dont think face shape is good / chins are in wrong direction
 

lewutt

DF Pleb
d33p3r said:
so I added some faces from extreme above angles to source and placed them into data_src/aligned. is it necessary/essential to have also the right names in the debug folder similar to those in aligned? I guess because I sorted somewhere before they do not match anymore.

Next again I deleted bad landscapes in data_dst/aligned_debug and manually re extracted these. I found pressing a button helps a lot to get the real extreme angles captchured correct this time. made a backup of my fst/aligned before, then added the new extracted faces to my backup. it overwrote some old files but again not all, I sorted out before I guess. I guess either now I have some bad ones that did not overwrote now.

hard to say, might redo dst completely. I will continue to train now for some more time lets see if it helps. settings as mentioned above since I might have more src than dst faces and pretty much angles in src I dont think random_flip will help and I will keep random warp enabled because I still dont think face shape is good / chins are in wrong direction

No matter what you change/delete/rename in any debug folder, it doesn't affect training in anyway. Debug folder is only a helpful visual for your eyes, if you wanna see how good your alignments are. Those green landmarks that you see in debug images are also embedded (but invisible) in the aligned folder - as long as you extracted faces with .bat command and not simply copy pasted some random faces that you downloaded in your aligned folder. Just keep in mind that every time you use extract faces .bat, everything in your aligned folder will be deleted. So if you wanna add more src/dst pics to your aligned folder, first you copy paste it somewhere safe, then you combine contents again. The aligned folder is all that matters. So don't worry about names in _debug. Keep in mind that if you are adding pics to dst aligned, name matters. Because when you merge for video, it is doing it frame by frame, based on name of aligned_dst. If you randomly extract more _dst from random parts of video, and they don't have the correct number as in video timeline, you will have random frames at random times when you merge/convert.

Also what do you mean pressing a button helped get the extreme angles? What button?
 

TMBDF

Moderator | Deepfake Creator | Guide maintainer
Staff member
Moderator
Verified Video Creator
lewutt said:
d33p3r said:
so I added some faces from extreme above angles to source and placed them into data_src/aligned. is it necessary/essential to have also the right names in the debug folder similar to those in aligned? I guess because I sorted somewhere before they do not match anymore.

Next again I deleted bad landscapes in data_dst/aligned_debug and manually re extracted these. I found pressing a button helps a lot to get the real extreme angles captchured correct this time. made a backup of my fst/aligned before, then added the new extracted faces to my backup. it overwrote some old files but again not all, I sorted out before I guess. I guess either now I have some bad ones that did not overwrote now.

hard to say, might redo dst completely. I will continue to train now for some more time lets see if it helps. settings as mentioned above since I might have more src than dst faces and pretty much angles in src I dont think random_flip will help and I will keep random warp enabled because I still dont think face shape is good / chins are in wrong direction

No matter what you change/delete/rename in any debug folder, it doesn't affect training in anyway. Debug folder is only a helpful visual for your eyes, if you wanna see how good your alignments are. Those green landmarks that you see in debug images are also embedded (but invisible) in the aligned folder - as long as you extracted faces with .bat command and not simply copy pasted some random faces that you downloaded in your aligned folder. Just keep in mind that every time you use extract faces .bat, everything in your aligned folder will be deleted. So if you wanna add more src/dst pics to your aligned folder, first you copy paste it somewhere safe, then you combine contents again. The aligned folder is all that matters. So don't worry about names in _debug. Keep in mind that if you are adding pics to dst aligned, name matters. Because when you merge for video, it is doing it frame by frame, based on name of aligned_dst. If you randomly extract more _dst from random parts of video, and they don't have the correct number as in video timeline, you will have random frames at random times when you merge/convert.

Also what do you mean pressing a button helped get the extreme angles? What button?

He meant A button you press to change accuracy mode in the manual extractor.
 
Top