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

[Propuscus]

Any have this??

You must be registered for see links

 

[Propuscus]

You must be registered for see images

What is the com2star number in this photo? Does anyone have one?

 

[cherryrin]

You must be registered for see links

Anyone know which Winter number this one is from C2S~??
I believe it's one of the newer ones?
I saw this on the Chinese site, but the link seems expired
Thanks~

 

[hannirani]

Alright guys, need little help from you all with finding p*rnstars that match the following idols i plan to make:

Haerin
Hanni (need both long and short haired)
Danielle
Minji
Minju from Illit (priority right now cause there is no minju DF out there)
Moka from illit (priority coz no DF of moka out there)
Natty from KIOF (priority)
Haneul from KIOF (priority)
Ningning Aespa (priority cause barely any DF of her out there)
Yunjin
Chaewon
Sakura
Kazuha


Post names or video code that you think will match any of the idols above. Really hoping less of JAV actresses and more western or russian or anyone that can match these idols. i hate JAV censored sh*t. so if you are suggesting JAV models or videos, make sure they have uncensored version. Cause i am mostly a BJ guy and plan to make mostly BJ videos with these idols. P.S. Minju from illit is really a priority right now

 

[snowbun]

i rarely request something, but do anyone have that karina with a good proportion with a red lingerie? i saw it somewhere, and i can't find it anymore. if anyone have it, i really appreciate u.

 

[YUNΛ]

MVE realeased new trainer for dfl that is over 2x faster and uses less vram than the old one!
If you have 8gb or more vram you are able to make better deepfakes than this

You must be registered for see links

(this is done with old dfl trainer and gpu with 8gb vram)

Spoiler: installation

you only need to install dfl

You must be registered for see links

if you have 3000 or 4000 or 5000 series nvidia gpu installing RTX 3000 series build

You must be registered for see links

if you have nvidia 2000 series you need intalling rtx 2000 build

You must be registered for see links

If you have amd (or nvidia 1000 series) you need to intalling direct x12 build

You must be registered for see links

after you have installed them you probably need installing mve fork (not 100% sure if it is needed but it is pretty useful)

You must be registered for see links

Click code button and download zip

Make new folder named deepfacelab and but the MVE fork files there.
then go to your deepfacelab and open file named _internal
move the file named deepfacelab (wich has the MVE fork files inside) there.



next you install the new trainer

You must be registered for see links

and but them into deepfacelab folder (not the on in the _internal)

now setup is ready!!!! (finally!)

Spoiler: setting up the model

next you need to make saehd model to train with
if you already have model you can use export_saehd_config.bat (you need to have adabelief enabled)
open the saehd_config.yaml (with text editor like notepad)

you should see list like this

general_params:
name: model
model_params:
archi: liae-ud
resolution: 128
ae_dims: 256
e_dims: 64
d_dims: 64
d_mask_dims: 22
face_type: wf
data_loading:
use_cache: true
image_type: jpg
image_quality: 1
force_recache: false
folder_suffix: .face_cache_
training_properties:
mp_mode: full_t
pin_samples: true
autobackup_hour: 3
maximum_n_backups: 24
batch_size: 8
batch_dst_mixing: 0.0
masked_training: true
eye_mouth_priority: false
blur_out_mask: true
uniform_yaw: true
lr: 5.0e-05
lr_dropout: false
cpu_cap: 6
clipgrad: false
keep_graph: false
loss: ssim
weight_decay: 0.0005
freeze_encoder: false
snapshot_interval: 5000
augmentations:
random_warp: true
random_noise: false
random_src_flip: false
random_dst_flip: true
color_augmentation: true
scale_range: 0.15
rotation_range: 2
tx_range: 0.05
ty_range: 0.05
gan:
gan_training: false
gan_power: 0.01
gan_dims: 16
gan_patch_size: 64
gan_weight_decay: 0.001
gan_smoothing: 0.3
gan_noise: 0.01


you should change atleast these thinks (idk what most of the thinks do cause they are new and I haven't testing yet)
you should check this

You must be registered for see links

for optimal res and dims for your gpu (only vram should matter)


general_params:
name: model (put your model name can be anything!)
model_params:
archi: liae-ud (you can try with diffrent model arc (i use liae-ud))
resolution: 128 (look what other people use with same ammount of vram)
ae_dims: 256 (should be same with res or higer)
e_dims: 64 (you should again check what other people use with same ammount of vram)
d_dims: 64 (this number has to be same with e_dims or model corrupts)
d_mask_dims: 22 (idk about this i use 28)
face_type: wf (just use wf)
data_loading:
use_cache: true
image_type: jpg
image_quality: 1
force_recache: false
folder_suffix: .face_cache_
training_properties:
mp_mode: full_t (change this to bfloat it should be faster)
pin_samples: true
autobackup_hour: 3
maximum_n_backups: 24
batch_size: 8 (this should be 8 but you can use 6-10 but other values are considering bad)
batch_dst_mixing: 0.0
masked_training: true
eye_mouth_priority: false (use if eyes look weird after long amount of training)
blur_out_mask: true ()
uniform_yaw: true
lr: 5.0e-05
lr_dropout: false (need to enable this on some point but not in start)
cpu_cap: 6 (change this so it matches with the amount of you cores in your cpu)
clipgrad: false (change to true)
keep_graph: false
loss: ssim
weight_decay: 0.0005
freeze_encoder: false
snapshot_interval: 5000
augmentations:
random_warp: true (should be true at the start but need dissable at the end of training)
random_noise: false
random_src_flip: false
random_dst_flip: true
color_augmentation: true
scale_range: 0.15
rotation_range: 2
tx_range: 0.05
ty_range: 0.05
gan:
gan_training: false (you can enable this at the end of training)
gan_power: 0.01
gan_dims: 16
gan_patch_size: 64
gan_weight_decay: 0.001
gan_smoothing: 0.3
gan_noise: 0.01

Spoiler: data_src and data_dst

now you can start training if you have some faceset on your workspaces data_src/aligned folder (faceset for person who you wanting to deepfake)

and some faceset in data_dst/aligned folder (destination video)

you can get data_dst files here

You must be registered for see links

or make your own

Spoiler: training

now you should be ready to start training
open train_saehd.bat and start training
you need to use your browser and go to

You must be registered for see links

there you have you preview window
now you just let it train and wait!!!
after you are done click exit (it will save it automatically)

 

[Useeeer1234]

원래의???

You must be registered for see images

 

[Useeeer1234]

Sorry about no photo problem

Screenshot 20250111 041155 Photos~2

Image Screenshot 20250111 041155 Photos~2 in Files by taeho album

www.imagepond.net

 

[Norway1]

Anyone with doge 777?

 

[mamangkesbor]

Is SSS dfs already there on the internet? Many of the members are old enough

 

[Sayakroco12]

Anyone know original number this

You must be registered for see links

 

[DegenDonkey]

heres what ive made thanks to @YUNΛ's faceset (thanks btw)

You must be registered for see links

still cooking other vids and versions, and i am once again begging for facesets... anyone wanna share?

 

[MinaLover37]

You must be registered for see links

anyone kind enough to give the full version. I'll try to repay if I can with some from my archives.

 

[klee141181]

Any roop tips? I'm a complete noob but am having decent results with just face pictures for source files and gifs for target (my PC isn't great). Definitely have room for improvement

 

[Saif09]

does anyone have this

You must be registered for see links

it's one of the oldest vids but top notch quality