Mr DeepFakes Forums
  • New and improved dark forum theme!
  • Guests can now comment on videos on the tube.
   
StrayNucleusDeepFaceLab - An Amateur's Guide
#1
#2
Wait why is their only an extract all frames for the destination source?

Is this the way you do it, train a model for just that destination video every time?

If I want to use the same destination and source actors for different videos, isn't it easier to train one time with lots of footage of both?

Extract from destination 10-15 fps, just to train. Then use the generated model, and proper steps to make a video with that.
#3
(02-05-2019, 07:29 AM)wbdragon Wrote: You are not allowed to view links. Register or Login to view.Wait why is their only an extract all frames for the  destination source?

Is this the way you do it, train a model for just that destination video every time?

If I want to use the same destination and source actors for different videos, isn't it easier to train one time with lots of footage of both?

Extract from destination 10-15 fps, just to train. Then use the generated model, and proper steps to make a video with that.

You will need to extract every frame anyways if you want to make the entire video. Yes, theoretically you can cut the dst video into 10-15 fps just to train, but then you'll have to run it again to collect every frame to convert.

If you just just a 30fps dst video into 15 fps, you'll be making a choppy deepfake video in the end.
#4
Could I do this. I edit destination model from many clips to one video. Insert that video as a source to extract all aligned faces 10 fps. Take the good ones and put them in the destination aligned faces, and train from their.

Do file names matter when training aligned faces?

Yes ok, I just figured out that you need those frame to generate video.

Edit: Ok so if this works, theoretically, I could only train once well. And use that model, insert new destination video, get all frames, align, morph frames, and generate video?Something like that, haven't walked through yet.

Edit: The .bat does 10fps for src. not 15 opps
#5
(02-05-2019, 07:39 AM)wbdragon Wrote: You are not allowed to view links. Register or Login to view.Could I do this. I edit destination model from many clips to one video. Insert that video as a source to extract all aligned faces 10 fps. Take the good ones and put them in the destination aligned faces, and train from their.

Do file names matter when training aligned faces?

Yes ok, I just figured out that you need those frame to generate video.

Edit: Ok so if this works, theoretically, I could only train once well. And use that model, insert new destination video, get all frames, align, morph frames, and generate video?Something like that, haven't walked through yet.

Edit: The .bat does 10fps for src. not 15 opps

Not when training but I am pretty sure they matter when converting your video. Inevitably, you'll have to extract destination frames again if you make any modifications.
#6
nice guide!
#7
hi..if i want to start working on deepfakelab on ubuntu..is there any documentation exist to go ahead..i have done same on windows but now want to move on ubuntu..i have setup enviornment as mentioned in linux fork...how to run commands...do i need to convert windows .bat files to ubuntu .sh files to go ahead ?
#8
Can someone explain what the difference is between DF and H128? Both swap and dont morph, both are 128px. So why are they two separate trainers and not one?
#9
(06-07-2019, 02:22 AM)appass Wrote: You are not allowed to view links. Register or Login to view.Can someone explain what the difference is between DF and H128? Both swap and dont morph, both are 128px. So why are they two separate trainers and not one?

One is a full face model, the other one (h128) is only half face.
#10
(06-07-2019, 02:24 AM)easy40 Wrote: You are not allowed to view links. Register or Login to view.
(06-07-2019, 02:22 AM)appass Wrote: You are not allowed to view links. Register or Login to view.Can someone explain what the difference is between DF and H128? Both swap and dont morph, both are 128px. So why are they two separate trainers and not one?

One is a full face model, the other one (h128) is only half face.

Lmao that explains why the lips were clipped on my h128 attempt and not on DF. I assume then that @You are not allowed to view links. Register or Login to view. is using DF? His videos are so darn good, if I only knew his secret.

Forum Jump:

Users browsing this thread: 1 Guest(s)