[SHARING] DFL 2.0 ReadyToTrain model

[tonetantillo]

could someone help me with this error pls ?

Right click on model folder and make sure Read ONly in UN checked

 

[DeliveryBoy]

cool. i am working on better RTM model based on AMP.

i just bought myself the RTX3090, cant wait to get yoru new AMP model   thank you sir.

 

[DeliveryBoy]

I am working on a 320 RTM female model based on Adult Performers Xseg'd WF DST files from Groggy4 with my new RTX 3090

My aims for this 320 Ready To Merge Model are:

2 million Pretraining,
1 million RW
1 million RW + LR Dropout
1 million LR Dropout + GAN

NOTE: i am also doing a 256 RTM version as well for those people who needed a model for Adult Performers.

 

[iperov]

Is there a posibility to make my own RTM model and use it in DeepFaceLive but as a head model instead of whole face?

yes but you have to train with particular dst face manually xseg masked

 

[TMBDF]

I urge everyone to not use this thread for asking questions, there is already a complete and detailed explanation and workflow for training RTM models in literally the first post "iperov" made by creating this thread, use it only to post models, if you have an unique but simple questions that can be answered with yes/no ask in shoutbox, for more complex questions head over to questions section and create a new thread or ask in the DFL discussion thread.

 

[jaklins12]

<p>i've trained the src model, but it always look like the dst model. I've trained for about 30k iterations but the face hasn't changed. What am I doing wrong?</p>

 

[DeliveryBoy]

Almost completed with 320 RTM, i have to do 300K GAN then the 320 RTM model will be finished and ready for you guys to use.

 

[liebesleid]

Almost completed with 320 RTM, i have to do 300K GAN then the 320 RTM model will be finished and ready for you guys to use.

Thanks for your work!

 

[DeliveryBoy]

Almost completed with 320 RTM, i have to do 300K GAN then the 320 RTM model will be finished and ready for you guys to use.

Thanks for your work!

You're welcome it should be finished in a couple of days when i reach 3,800,000 iterations.


For those that can't wait for the 320 RTM to reach 3,800,000 iterations here is the download link for the 320 RTM model that is very close to 3,700,000 iterations.

Download Link:

 

[honor]

Almost completed with 320 RTM, i have to do 300K GAN then the 320 RTM model will be finished and ready for you guys to use.

Thanks for your work!

"I am working on a 320 RTM female model based on Adult Performers Xseg'd WF DST files from Groggy4 with my new RTX 3090

My aims for this 320 Ready To Merge Model are:

2 million Pretraining,
1 million RW
1 million RW + LR Dropout
1 million LR Dropout + GAN

NOTE: i am also doing a 256 RTM version as well for those people who needed a model for Adult Performers. "

I presume you are following the earlier workflow you had mentioned.

If so, you have already done :

- 2M PreTraining
- 1M RW
- 1M RW + LRD

and you are doing the last step in your workflow : 1M LRD + Gan - ja ?

Thanks for the massive effort.. and sharing..

I was wondering looking at those loss numbers, why your Src loss is half ( or lesser than that ) of your Dst Loss ?

at Src end - Can you mention what is the celeb/ Src ( FaceSet ) you are using in your training ?
at Dst end - Also, what Dst ( Faceset ) are you using ? ( is it groggy's Famous training Set )

 

[tenorf]

Almost completed with 320 RTM, i have to do 300K GAN then the 320 RTM model will be finished and ready for you guys to use.

Thanks for your work!

You're welcome it should be finished in a couple of days when i reach 3,800,000 iterations.

For those that can't wait for the 320 RTM to reach 3,800,000 iterations here is the download link for the 320 RTM model that is very close to 3,700,000 iterations.

Download Link:

Thank you very much, your work is amazing. I hope to be able to buy a powerful graphic in the future and to be able to contribute as well.

 

[TIDF]

For those that can't wait for the 320 RTM to reach 3,800,000 iterations here is the download link for the 320 RTM model that is very close to 3,700,000 iterations.

Download Link:

Amazing work, keep it up! Been using your 256 res LIAE RTM for a while now, it's so easy to make decent deepfakes quickly.

 

[DeliveryBoy]

You're welcome

 

[androsk]

Almost completed with 320 RTM, i have to do 300K GAN then the 320 RTM model will be finished and ready for you guys to use.

Thanks for your work!

You're welcome it should be finished in a couple of days when i reach 3,800,000 iterations.


For those that can't wait for the 320 RTM to reach 3,800,000 iterations here is the download link for the 320 RTM model that is very close to 3,700,000 iterations.

Download Link:

access needed

 

[synth17]

"DeliveryBoy" Why do you recommend turning off blur out mask? Should we turn off if training our own RTM? Or just turn off when merging with a new dst?

 

[ikke21]

This is my first attempt at creating an RTM model, so bear with me. It's a model for Phoebe Tonkin, an actress of moderate fame and great beauty. Here is the link: 

   (Let me know if it works, I've never used mega before.)

The model is trained for a bit more than 5.3 kk iterations. It is more or less according to the instructions by , so LIAE-UD, 224, [font='Open Sans', sans-serif]WF, ae_dims:256, e_dims:64, d_dims:64, d_mask_dims 22. It is trained on a destination set of my own concoction, NSFW. Here's the rough outline of the training:[/font]

1) About 2kk training with both random src and dst. lrd:N, uniform_yaw:True, blur_out_mask:Y, color_transfer:lct, train, eyes_mouth_prio:Y. Random src also my own concoction. I was experimenting with both src and dst during the first 1kk. After that, I settled on adding the dst images to the src folder also. This was the only way I could get the eye directions right.

2) About 1kk with the same settings, but with only Tonkin in the src.

3) About 500k with lrd:Y.

4) About 500k with uniform yaw: false.

5) About 1kk with random warp disabled. I realised somewhere halfway through that I was displeased with the way open mouths looked (in the "merge", column 5). If a mouth was open in dst and the tongue visible inside, this would often be mapped as a more closed mouth, with the tongue becoming a lower lip. I had initially masked the dst according to the main guide, so excluding tongues in open mouths in dst. I decided to retrain the xseg mask to include the tongue and teeth. For the second half of this stage I trained like that and it improved the open mouths, as best I could tell. Next time I will do this from the start. I would still use the masking from the guide for the merging stage.

6) About 150k gan 0.1 gan_dims:32.

I would appreciate any feedback.

 

[DeliveryBoy]

i saw message i got banned so i disabled access to my model

I got really ugly result when i enabled blur out mask.

 

[yonff]

If I were to create a custom aligned dst set/pretrained faceset for my own RTM model how many aligned images should it be?

 

[TMBDF]

This is my first attempt at creating an RTM model, so bear with me. It's a model for Phoebe Tonkin, an actress of moderate fame and great beauty. Here is the link: 

   (Let me know if it works, I've never used mega before.)

The model is trained for a bit more than 5.3 kk iterations. It is more or less according to the instructions by , so LIAE-UD, 224, [font='Open Sans', sans-serif]WF, ae_dims:256, e_dims:64, d_dims:64, d_mask_dims 22. It is trained on a destination set of my own concoction, NSFW. Here's the rough outline of the training:[/font]

1) About 2kk training with both random src and dst. lrd:N, uniform_yaw:True, blur_out_mask:Y, color_transfer:lct, train, eyes_mouth_prio:Y. Random src also my own concoction. I was experimenting with both src and dst during the first 1kk. After that, I settled on adding the dst images to the src folder also. This was the only way I could get the eye directions right.

2) About 1kk with the same settings, but with only Tonkin in the src.

3) About 500k with lrd:Y.

4) About 500k with uniform yaw: false.

5) About 1kk with random warp disabled. I realised somewhere halfway through that I was displeased with the way open mouths looked (in the "merge", column 5). If a mouth was open in dst and the tongue visible inside, this would often be mapped as a more closed mouth, with the tongue becoming a lower lip. I had initially masked the dst according to the main guide, so excluding tongues in open mouths in dst. I decided to retrain the xseg mask to include the tongue and teeth. For the second half of this stage I trained like that and it improved the open mouths, as best I could tell. Next time I will do this from the start. I would still use the masking from the guide for the merging stage.

6) About 150k gan 0.1 gan_dims:32.

I would appreciate any feedback.

You know you can use any other resolution? 224 is kinda low, especially for WF, it's equivalent of like 160FF, unless your GPU is a limiting factor you really shouldn't be using 224 becasue you are not gonna get very detailed results, max out your GPU.

 

[up_side_down]

i saw message i got banned so i disabled access to my model

I got really ugly result when i enabled blur out mask.

The admins of this site would Ban you for sharing Your RTM?