Fakes - Kpop Deepfakes (Com2star, MondoKpop, OnceFake, Etc) | Page 129 | Social Media Girls

Fakes Kpop Deepfakes (Com2star, MondoKpop, OnceFake, Etc)

Update

I've finished both Haerin and Sakura albums


I'm now going to do a IU album which I'll upload bit by bit until it's fully updated with and will do another Idol next when I eventually post again.


And as always I'm uploading daily on my main album


Where I'm currently uploading mainly captain_y videos (eventually will upload gwfakes videos and other creators) and the uploads start from old to new videos from the creator.

That's all
 
S
seely commented
U
UltimateK123 commented
Yooa please 🙏
 
datlmhg1
datlmhg1 commented
Jessica, Jiyeon pls
 
B
bibibibilong commented
Yooa plszzz
 

tempmax

New member
Oct 3, 2024
2
31
I'm working on creating a DFL faceset and wanted to hear how you all approach it. I've seen a lot of different methods, and everyone seems to have their own way of doing it.

I'm preparing for training at 512res or higher, so I'd love some advice on the best way to extract a faceset for that.

I've gone through the guides and tried making a faceset myself, but I'm wondering if there's a more efficient way to do it.

Thanks for any advice!
 
4 Comments
YUNΛ
YUNΛ commented
Don't really know what you are asking but here is something maybe helpful?

You need to find good quality videos (preferably 4k) of you Idol that also has good variety of different angles...
kpop On spotify is good channel all their videos are high quality + usually has many angles.

for video downloading use https://jdownloader.org/jdownloader2 or https://github.com/yt-dlp/yt-dlp
jdownloader is easier to use, but yt-dlp is faster. (downloading a 1gb of a video takes like 1min with jdownloader vs 10s with yt-dlp)

You can also use single images in src to get more angles. good sites for high quality images are kpopping and pinterest

you try to get as many angles as possible you can use mve's https://drive.google.com/file/d/1nu7Q7cVvKUkZUeRP8Owtw6_43UZyTZy4/view heatmap tool to check what angles you have and what are missing.
(more squares is better)


for deleting duplicate images you can use tools like dupeguru or visipics.

Idk if 512res models need somehow different type of scr sets, but onlyfake30 who trains with 512 model also uses 512 res src set (unless he resized it to smaller size for sharing...)
 
O
onlyfake30 commented
As far as I know, no matter how much size you modify the image (even 2048*2048), the DFL will still reduce or enlarge the image size by the CPU during training to match the res of your model. so 512res model also uses 512*512 SRC,In fact, if you have a CPU above 12700K, 128res to 512res models can uniformly use 512*512 SRC, cause the CPU can change the resolution of the image to match the model at a negligible time
 
YUNΛ
YUNΛ commented
I think you are right about training but it still makes some difference (at least on dst). here is the same video other one is 512x512 and the other 1024x1024 (trained 1000 iterations longer). the setting used in Merger are the same, but I think you can quite easily tell which one is the 512x512 version
https://media.imagepond.net/media/512-vs-102415265f32cc1284c1.mp4
 
T
tempmax commented
So, would it be fair to say that there's not much difference between 512*512 and 1024*1024?
 
anyone here have experience with PixVerse prompts? trying to bypass nsfw moderation
 
  • Like
Reactions: 3dvr
1 Comment
pokekemaster
pokekemaster commented
This is what i have. just to remove top cloth:

The person pictured does not stand up and does not change their expression or the direction they are looking at all. Her face does not change expression and remains prominently in view. Her eyes remain open. Her face stays in shot. She does not turn her head. The camera movement is dynamic.

With her gaze fixed firmly on the camera, she reaches toward the top portiong of her upper garment. Her hands move with precision and control, gripping the fabric deliberately. Without glancing at her hands or breaking her focus, she begins to pull the garment open from the top. The fabric strains before it removes completely, the sound of the moving fabric adding intensity to the moment. As the garment is pulled all the way down fully open, it falls away, revealing her torso and she keeps her gaze fixed firmly on the camera. Soft, warm lighting highlights the smoothness of her skin and the natural curves of her body, emphasizing her exposed tiny realistic bust and tiny areola with grace and subtlety.

After tearing the garment, she places her arms gently by her sides, her posture calm and composed. The camera continues to focus on her face, capturing the serenity and stillness in her expression. The rise and fall of her breath adds a quiet rhythm to the scene, while the camera remains focused on her face, maintaining the balance between stillness and subtle motion.
 
Kinda disappointed that this is the only Fromis9 I could find.
I added some extra albums to free up storage and because I won't be uploading for a few days until my data limit eventually reset.
 
R
rkwoaos23 commented
J
Jacey17 commented
thx for JYR & HW goat
 
M
magicfantafloat commented
sure I will upload some later maybe on my next post
sorry I'm only on C2S
 
R
rkwoaos23 commented