Fake In roop unleashed how do I render a good BJ + can you give me some general recommendation?

  • Welcome to the Fakes / AI / Deepfakes category
    When posting AI generated content please include prompts used where possible.
Sort By Date Sort By Reaction Score
simpilo

simpilo

Tier 3 Sub
May 8, 2023
25
258
477
0fya082315al84db03fa9bf467e3.png
I've seen deepfake videos where the blowjobs are decenltly managed, but I imagine it's difficult.

What are some general advice to make good videos? I do post-processing with GFPGAN and I also can manage when the cock is in front of the face using masks but I don't fully understand why it works.

Are there some guides that are specific about this stuff?

Also is there a way to make the swapped faced more expressive, because the eyebrows don't move at all in my video and it's weird.

Thank you for any suggestion you can provide


EDIT/UPDATE: so after some trial and error, and especially trying the most obvious thing. Using clip2seg I added words like cock, blowjob, lips wrapped around cock, tongue and processing time skyrockets but the results is a lot better. Even the face gets more expressive somehow.
The time to buy a decent graphic card gets closer and closer I guess
 
Last edited:
  • Like
Reactions: sdkmci, mkswolft, Zaynish and 4 others
mkswolft

mkswolft

Bathwater Drinker
Mar 12, 2022
281
8,671
1,308
0fya082315al84db03fa9bf467e3.png
simpilo said:
Please, Log in or Register to view quotes
Click to expand...

thanks for the update! Just wanna know more about roop unleashed. are you making the videos with just one or a few reference images? Or did you use Face Management to create a .fsz file to always have the face ready?

If you made a .fsz file, how many images did you use and which ones gave you the best results? Like, for creating a Lora, I know you need to train the images a certain number of times. If I give it a batch of 50 faces, will it turn out better than using 200? or if I use a famous model with 1500+ images from Instagram, would that be better, or is that overkill? or is just one image enough?

btw, another option I see is opening a video editor like Premiere and overlaying the roop video track on the original, then using a soft mask to keep the original mouth frame by frame to avoid face distortions. but yeah, that takes more time, maybe less than using clip2seg though. How much time difference did you notice between clip2seg and DFL Xseg? Haven’t tried clip2seg yet.

and for "Max Face Similarity Threshold," what value do you use? It’s set to 0.65, but I don’t see much difference when I set it to 0 or 1. Is it just me or are the results the same? seriously, I don’t see any difference lol

My best settings so far:
  • Blur size: 50
  • Face masking engine: DFL Xseg
  • Post-processing: GFPGAN
  • Original/Enhanced image blend ratio: 1
 
abrahamford1

abrahamford1

Tier 1 Sub
Apr 5, 2022
12
108
387
0fya082315al84db03fa9bf467e3.png
mkswolft said:
Please, Log in or Register to view quotes
Click to expand...
wans't for me the question but I can add that me for example, i use just one face/picture, the ones that i have more sucess are the ones that it's a front selfie of course but you can see a bit of the sides of the face, like ~30Âş angle pic, plus the face format needs to match at least a bit (egg shapped heads for me are the worst to deal with, nothing fits), I didn't got good results with face Management. and I read in the forums that the similarity threshold is more about multiple faces in the video, something like that, in other words, not that relevant.
 
  • PeepoLove
Reactions: mkswolft
You must log in or register to reply here.
Top Bottom