Newbie - should I continue with this data/training

Want to understand the training process better? Got tips for which model to use and when? This is the place for you


Forum rules

Read the FAQs and search the forum before posting a new topic.

This forum is for discussing tips and understanding the process involved with Training a Faceswap model.

If you have found a bug are having issues with the Training process not working, then you should post in the Training Support forum.

Please mark any answers that fixed your problems so others can find the solutions.

Post Reply
User avatar
Gaben1994
Posts: 1
Joined: Sat Mar 02, 2024 12:52 am
Has thanked: 1 time

Newbie - should I continue with this data/training

Post by Gaben1994 »

Hi everyone, young player here, I try to create a deepfake video to my friend's birthday, I am getting very close to the deadline so would like to ask experienced players if I should give up below training after 135 000 iterations (villian trainer)? I have this at the moment after that many iterations, probably I should have shaved and maybe I should wear a hat:

If I should give up, could you give me a few suggestions how to improve the new training? Since I am new, I don't even know what information should I share that could help people here with suggestions. Btw at the moment the average loss is: face_a: 0.02814, face_b: 0.01728.

Last edited by Gaben1994 on Sat Mar 02, 2024 1:09 am, edited 1 time in total.
User avatar
torzdf
Posts: 2679
Joined: Fri Jul 12, 2019 12:53 am
Answers: 159
Has thanked: 133 times
Been thanked: 625 times

Re: Newbie - should I continue with this data/training

Post by torzdf »

Villain is a notoriously slow model to train. so you may want to switch to something like dfaker to get quicker results (at slightly lower quality).

Note that neither of these models will be good for close-up shots as they output at 128px. Bigger models require more VRAM and more time to train though.

Also, see here: I've trained the model for ages, the previews look good, so why is the swapped face blurry?

My word is final

User avatar
d3x
Posts: 9
Joined: Tue Mar 12, 2024 1:12 pm
Has thanked: 2 times
Been thanked: 2 times

Re: Newbie - should I continue with this data/training

Post by d3x »

135k iterations is nothing 😅 Seems very undertrained
With villain (or any trainer with high resolution output) it always takes me a couple million iterations before it gets all the details right

Sidenote: when converting you can increase the Face Scale in the Face Processing section to better line up the edge of the replacement face with the original face (I usually have to increase mine by at least 1 or 2 to make it fit good)

But as mentioned above: with only 128px output, close-ups will always be kind of blurry - decreasing the Output Scale for the converted image or video can help make this less noticeable (compared to everything outside the swap area)

Last edited by d3x on Fri Apr 05, 2024 6:50 pm, edited 7 times in total.
Post Reply