[SHARING] DFL 2.0 ReadyToTrain model

[iperov]

"dpfks"
^ that shit should be deleted from my thread

 

[DeliveryBoy]

-----
Suggest description.
-----


ReadyToMerge model can be used to swap the face without the training.

But you can continue training with the same parameters on the new dst and improve the quality.

----------------
ReadyToMerge LIAE-UD model can be reused to train new fake from scratch.
All you need is delete INTER_B.npy file from the model folder. Also don't forget to set training settings to initial, such as RW(ON), LRD(off), GAN(0.0), etc... Don't enable/disable pretraining !

The generalization with poor dst faceset will be better than training with default pretrained model.

Example using Tom Cruise model to train Keanu Reeves fake on poor dst faceset from the game:

----------------

Download: <a href='/literotica/magnet:?xt=urn:btih:dfa9f27340097eb47c0c3cb9e7a8cb6d9876580a&dn=DeepFaceLabReadyToMergeModels&tr=udp%3a%2f%2ftracker.opentrackr.org%3a1337%2fannounce'>magnet link</a>


available celebs:

Tom Cruise (224, liae-ud)
[img=400x276]

How to create RTM model from scratch.

1) src faceset should be diverse enough in yaw, light and shadow conditions

src faceset should be xseg'ed and applied.

2) use default pretrained liae-ud model

3) use dst faceset from the torrent

4) train with:

res : 224
batch : 8
archi : liae-ud
face type : WF
random_flip : off
adabelief : on
eyes_mouth_prio : on
ct_mode: lct .
to 600k iters

5) enable LRD + 300k iters

6) disable random warp + 300k iters

7) enable gan 0.1 + 300k iters

This method is for swapping SRC or DST?

 

[DeliveryBoy]

-----
Suggest description.
-----



How to create RTM model from scratch.

1) src faceset should be diverse enough in yaw, light and shadow conditions

src faceset should be xseg'ed and applied.

2) use default pretrained liae-ud model

3) use dst faceset from the torrent

4) train with:

res : 224
batch : 8
archi : liae-ud
face type : WF
random_flip : off
adabelief : on
eyes_mouth_prio : on
ct_mode: lct .
to 600k iters

5) enable LRD + 300k iters

6) disable random warp + 300k iters

7) enable gan 0.1 + 300k iters

ct_mode: lct <--- this makes my model blue screen should i end training, the blue screen looked like the model messed up when i use LCT for CT_Mode.

 

[DeliveryBoy]

-----
Suggest description.
-----


ReadyToMerge model can be used to swap the face without the training.

But you can continue training with the same parameters on the new dst and improve the quality.

----------------
ReadyToMerge LIAE-UD model can be reused to train new fake from scratch.
All you need is delete INTER_B.npy file from the model folder. Also don't forget to set training settings to initial, such as RW(ON), LRD(off), GAN(0.0), etc... Don't enable/disable pretraining !

The generalization with poor dst faceset will be better than training with default pretrained model.

Example using Tom Cruise model to train Keanu Reeves fake on poor dst faceset from the game:

----------------

Download: <a href='/literotica/magnet:?xt=urn:btih:dfa9f27340097eb47c0c3cb9e7a8cb6d9876580a&dn=DeepFaceLabReadyToMergeModels&tr=udp%3a%2f%2ftracker.opentrackr.org%3a1337%2fannounce'>magnet link</a>


available celebs:

Tom Cruise (224, liae-ud)
[img=400x276]

How to create RTM model from scratch.

1) src faceset should be diverse enough in yaw, light and shadow conditions

src faceset should be xseg'ed and applied.

2) use default pretrained liae-ud model

3) use dst faceset from the torrent

4) train with:

res : 224
batch : 8
archi : liae-ud
face type : WF
random_flip : off
adabelief : on
eyes_mouth_prio : on
ct_mode: lct .
to 600k iters

5) enable LRD + 300k iters

6) disable random warp + 300k iters

7) enable gan 0.1 + 300k iters

Do i really need to do ct_mode: LCT? i am getting really funky red and sometimes blue all over the 5 face columns.

 

[mymisfake]

-----
Suggest description.
-----


ReadyToMerge model can be used to swap the face without the training.

But you can continue training with the same parameters on the new dst and improve the quality.

----------------
ReadyToMerge LIAE-UD model can be reused to train new fake from scratch.
All you need is delete INTER_B.npy file from the model folder. Also don't forget to set training settings to initial, such as RW(ON), LRD(off), GAN(0.0), etc... Don't enable/disable pretraining !

The generalization with poor dst faceset will be better than training with default pretrained model.

Example using Tom Cruise model to train Keanu Reeves fake on poor dst faceset from the game:

----------------

Download: <a href='/literotica/magnet:?xt=urn:btih:dfa9f27340097eb47c0c3cb9e7a8cb6d9876580a&dn=DeepFaceLabReadyToMergeModels&tr=udp%3a%2f%2ftracker.opentrackr.org%3a1337%2fannounce'>magnet link</a>


available celebs:

Tom Cruise (224, liae-ud)
[img=400x276]

How to create RTM model from scratch.

1) src faceset should be diverse enough in yaw, light and shadow conditions

src faceset should be xseg'ed and applied.

2) use default pretrained liae-ud model

3) use dst faceset from the torrent

4) train with:

res : 224
batch : 8
archi : liae-ud
face type : WF
random_flip : off
adabelief : on
eyes_mouth_prio : on
ct_mode: lct .
to 600k iters

5) enable LRD + 300k iters

6) disable random warp + 300k iters

7) enable gan 0.1 + 300k iters

Sorry, I'd like to try this...instructions for training are pretty clear, I just wanna ask this: should eyes_mouth_prio stay always enabled?

 

[DeliveryBoy]

gan_patch_size: 3
gan_dims: 4

these are the only settings that will let me train without 00M error

Are these settings a waste of time, will i see pores pimples and facial hair come through after merging?

My src has facial hair.

 

[DeliveryBoy]

OMFG i didn't train look what ReadyToMerge Models could do without training new DST???

 

[Ghostrox]

I have this weird bug, the face of scr is not getting aligned with the dst. I tried many times. Some help, please!

 

[jmaande]

"iperov", What model do you mean with the "default pretrained liae-ud model" in the second step of creating a RTM model from scratch? Is this the example Tom Cruise model you made...

 

[Cioscos]

"iperov", What model do  you mean with the "default pretrained liae-ud model" in the second step of creating a RTM model from scratch? Is this the example Tom Cruise model you made...

Just use a pretrained model with the res and ae, e and d which u desire.

---

I'm creating a rtm model with higher res. I will share the updates.

 

[DeliveryBoy]

"iperov", What model do  you mean with the "default pretrained liae-ud model" in the second step of creating a RTM model from scratch? Is this the example Tom Cruise model you made...

Just use a pretrained model with the res and ae, e and d which u desire.

---

I'm creating a rtm model with higher res. I will share the updates.

---

May i have resolution 256 liae-ud wf please?

 

[Cioscos]

"iperov", What model do  you mean with the "default pretrained liae-ud model" in the second step of creating a RTM model from scratch? Is this the example Tom Cruise model you made...

Just use a pretrained model with the res and ae, e and d which u desire.

---

I'm creating a rtm model with higher res. I will share the updates.

---

May i have resolution 256 liae-ud wf please?

---

Mine will be 320 res and it means half month or more of Google Colab. So if u pay me yes

 

[DeliveryBoy]

"iperov", What model do  you mean with the "default pretrained liae-ud model" in the second step of creating a RTM model from scratch? Is this the example Tom Cruise model you made...

Just use a pretrained model with the res and ae, e and d which u desire.

---

I'm creating a rtm model with higher res. I will share the updates.

---

May i have resolution 256 liae-ud wf please?

---

Mine will be 320 res and it means half month or more of Google Colab. So if u pay me yes

---

What does this mean Google Colab?

 

[Cioscos]

"DeliveryBoy" It's a platform which allow you to run cloudly a jupyter notebook. Chervonij created one of them on Colab with all commands to execute DFL there. It has Free and Pro account.

 

[nekouu]

Thank you very much for sharing this model.
It worked fine for me, but I just curious what are the roles of the model files.
For example, why we should delete INTER_B.npy?
May someone discuss some of the principles here?

 

[DeliveryBoy]

So every time we use new DST or new SRC we just delete INTER_B.npy ?

 

[texapo9]

This will be my first real pre-train model from scratch, but in next day or so I should have a 320 res 256/64/22/22 LIAE-UD pre-train model iterations available to share, depending on returns might stop at 500K. For anyone that needs it (I'll post it to the pre-train thread as well)

 

[Dim89]

>ReadyToMerge LIAE-UD model can be reused to train new fake from scratch.
>All you need is delete INTER_B.npy file from the model folder. Also don't forget to set training settings to initial, such as RW(ON), LRD(off), GAN(0.0), etc... Don't enable/disable pretraining !

Is it possible to change face_type option (from wf to head) of this saedh when i am training new dst?

 

[texapo9999]

320 Res, LIAE-UD Pre-Train Model, using Groggy's Female Faceset, will restart training on this to 1M in another week or two, had to pause. Working towards this thread for the RTM Model for a specific faceset. First pre-train model from scratch so if there's anything wrong with the dims my bad
resolution: 320
face_type: wf
archi: liae-ud
ae_dims: 256
e_dims: 64
d_dims: 64
d_mask_dims: 22
Current Iterations: 573,200

 

[etsnotes]

Magnet link of Iperov model is dead. Could you please share a new link? Many thanks.