Mr DeepFakes Forums
  • New and improved dark forum theme!
  • Guests can now comment on videos on the tube.
   
cazakavgDFL Questions - New User Help
#1
Dear Community,

Sorry for the title being so vague, my questions are quite broad and varied so I could not come up with a concise enough title.

I installed DeepFaceLab yesterday and have followed a few tutorials on how to use it. I would classify myself as a beginner-Amateur now (I have spent too many hours over the past two days training ahaha). I have a Ryzen 2700X and a RTX 2070 with 16GB 2666mHz RAM (this is getting upgraded to 48 GB soon).


So onto the questions :

Is there a guide which is a "fixing mistakes" type of thing? On "DeepFaceLab Explained and Usage Tutorial" dpfks explains what each of the 'sliders' do for convert options and training. Is there a guide or a video where a series of pictures are shown and what is wrong with them (i.e. why they do not look amazing despite tonnes of training) and how to fix them? (e.g. change face scale by x value). I have looked at the examples of the extremes dpfks has given but I am struggling to apply that to my trainings.

Can src models be reused for other dst models and vice versa? If I have spent a day training Emma Watson and Jade Nile, can I extract the second "pictures" from the training preview to be used for other models? In my eyes, having to repeatedly retrain (in this example) Emma Watson's face to be used with different dst actresses is a needlessly repetitive process. Surely there is a way to save that model of Emma Watson (and the dst actress) so that they can be mixed and matched with other src and dst models saving a tonne of time remodelling in the future. I hope I have explained this point clearly, I apologise if this is not the case.

In the case of facial cumshots, is there a particular method which can be used to allow the ejaculate to be shown on the exported face? (rather than distorting the image). The same applies for other blockages such as appendages as they have only served to mess up the face merge for me.

What does continuing training with different settings achieve? Is there a specific purpose for this or should one simply pick the best settings upon the initialisation of the training to achieve the best modelling outcome?

What is the optimum method for achieve adult actresses faces. I have read that for celebrities, interviews on YouTube are the way to go. I do not have an account for any pornographic websites and would prefer not to sign up to any, so is there a free, non-account required website that is reliable for downloading the thousands of high quality images needed for DFL training?


Thank you to anyone who is able to answer the above. It is most appreciated. 

Also. if anyone is desiring a particular model, feel free to message me and I will see what I can do as my computing power is here to serve the community. If you could assist with any of the process, i.e. what settings I should use or photo banks of the specified model, that would be great.
#2
I can answer the last question. 

You can always use torrent files to get adult content, but I advise using a VPN when doing so. (And I would suggest a paid VPN, at that.) You can use piratebay, rarbg, kickasstorrents, or 1337x for searching videos.

You can register for a free account at planetsuzy.com. You can search literally thousands of HD videos for your actress. The downloads will be slow, however, unless you wish to pay.

Another option is to get a video downloader extension for Firefox or Chrome (your choice). Then do a Bing.com search for HD videos of your actress, and use the extension to rip the video.

I have used all three successfully.
#3
(04-16-2019, 04:30 PM)cazakavg Wrote: You are not allowed to view links. Register or Login to view.Dear Community,

Sorry for the title being so vague, my questions are quite broad and varied so I could not come up with a concise enough title.

I installed DeepFaceLab yesterday and have followed a few tutorials on how to use it. I would classify myself as a beginner-Amateur now (I have spent too many hours over the past two days training ahaha). I have a Ryzen 2700X and a RTX 2070 with 16GB 2666mHz RAM (this is getting upgraded to 48 GB soon).


So onto the questions :

Is there a guide which is a "fixing mistakes" type of thing? On "DeepFaceLab Explained and Usage Tutorial" dpfks explains what each of the 'sliders' do for convert options and training. Is there a guide or a video where a series of pictures are shown and what is wrong with them (i.e. why they do not look amazing despite tonnes of training) and how to fix them? (e.g. change face scale by x value). I have looked at the examples of the extremes dpfks has given but I am struggling to apply that to my trainings.

Can src models be reused for other dst models and vice versa? If I have spent a day training Emma Watson and Jade Nile, can I extract the second "pictures" from the training preview to be used for other models? In my eyes, having to repeatedly retrain (in this example) Emma Watson's face to be used with different dst actresses is a needlessly repetitive process. Surely there is a way to save that model of Emma Watson (and the dst actress) so that they can be mixed and matched with other src and dst models saving a tonne of time remodelling in the future. I hope I have explained this point clearly, I apologise if this is not the case.

In the case of facial cumshots, is there a particular method which can be used to allow the ejaculate to be shown on the exported face? (rather than distorting the image). The same applies for other blockages such as appendages as they have only served to mess up the face merge for me.

What does continuing training with different settings achieve? Is there a specific purpose for this or should one simply pick the best settings upon the initialisation of the training to achieve the best modelling outcome?

What is the optimum method for achieve adult actresses faces. I have read that for celebrities, interviews on YouTube are the way to go. I do not have an account for any pornographic websites and would prefer not to sign up to any, so is there a free, non-account required website that is reliable for downloading the thousands of high quality images needed for DFL training?


Thank you to anyone who is able to answer the above. It is most appreciated. 

Also. if anyone is desiring a particular model, feel free to message me and I will see what I can do as my computing power is here to serve the community. If you could assist with any of the process, i.e. what settings I should use or photo banks of the specified model, that would be great.

1) No guide yet for fixing mistakes. Just post errors as they come up, if there are enough examples maybe we can compile them into a guide.
2) Yes you can re-used the trained models with caution. Re-using models increases the risk of the model collapsing/corruption. The general practice is to change one faceset at a time.
3) In general no, masking those obstructions cannot be done currently. There is an experimental masking feature in the newest DFL version, where you literally have to go through each frame and mask out any obstructions from each frame.
4) Not sure what you mean by "continuing training with different settings". You can always modify batch size, optimizer, style powers, and pixel loss. No right or wrong way to train, but generally pixel loss is no longer recommended, or I personally recommend turning on only after 100k iterations with backups on.
5) YouTube is my #1 source for building a faceset. Movies or TV shows #2. As for adult videos, you can download off PornHub (sign up required), or I generally torrent.

Forum Jump:

Users browsing this thread: 1 Guest(s)