[Newbie] Train with another source

Want to understand the training process better? Got tips for which model to use and when? This is the place for you


Forum rules

Read the FAQs and search the forum before posting a new topic.

This forum is for discussing tips and understanding the process involved with Training a Faceswap model.

If you have found a bug are having issues with the Training process not working, then you should post in the Training Support forum.

Please mark any answers that fixed your problems so others can find the solutions.

Locked
User avatar
slitchaser
Posts: 4
Joined: Thu Mar 23, 2023 8:45 am
Has thanked: 1 time

[Newbie] Train with another source

Post by slitchaser »

*** Please forgive me for my grammar ***
Hi , I'm new here. I have done covert some of simple videos before . But there are something that I curious. I will tell my situation first, what I have done and what I want to do next.(If something wrong plz tell me)

I have done

  • Video_A > Extract > RemoveFace > Update Alignment
  • Video_B > Extract > RemoveFace > Update Alignment
  • Train with DFL-H128
  • Convert(B>A) > Result1

What I want to do next ...
-I have Video_C which have same actor with Video_A.
-Video_C > Extract > RemoveFace > Update Alignment
-I want to swap B>C.

Can I use model from previous session(B>A) without training ? Or Any suggestion ?

Other question
does extract face contain any data ? I have read some topic here and someone not sure mention about update png? or something like that. plz tell me more about it.

Do change setting while training effect immediately or have to stop and training again?

About Mask , Do I understand correct ? In Extract we can have multiple mask but we can Train only one mask at a time. If I select Bisnet-FP , Vgg obstructed and Unet-dfl Masker. In alignment file will contain all 3 masker data right ? . In training I can train in each masker right ?
Thx.

Last edited by slitchaser on Thu Mar 23, 2023 10:39 am, edited 4 times in total.
User avatar
bryanlyon
Site Admin
Posts: 793
Joined: Fri Jul 12, 2019 12:49 am
Answers: 44
Location: San Francisco
Has thanked: 4 times
Been thanked: 215 times
Contact:

Re: [Newbie] Train with another source

Post by bryanlyon »

When you train a model you're training it on two faces, not two videos. You can then use it to swap those two faces in any videos. If you are using an unbalanced model such as Unbalanced or Reface then you'll get the best results putting the B face onto the A video, other models do both directions equally but you'll have to enable "swap models" to put the A face on the B video.

Extracted faces include all the information necessary for training including the face information and the mask. The data is stored in the PNG metadata.

Some training settings can be changed, but settings such as mask or loss functions are set when training begins and not updated even if you change it later.

Training can use a mask, but it does not train the mask unless youenable "learn mask" (which is probably not what you want). My suggestion is to just use the Bisenet mask. It's by far the best.

User avatar
torzdf
Posts: 2636
Joined: Fri Jul 12, 2019 12:53 am
Answers: 156
Has thanked: 128 times
Been thanked: 614 times

Re: [Newbie] Train with another source

Post by torzdf »

bryanlyon wrote: Thu Mar 23, 2023 5:34 pm

Some training settings can be changed, but settings such as mask or loss functions are set when training begins and not updated even if you change it later.

Just a clarification on this, loss functions can be changed. However, settings are generally locked to a model. Settings which can be adjusted for existing models are clearly labelled in the tooltips.

My word is final

Locked