It's Justin Timberlake, like, there isn't much to talk about. Nsync, solo career... actor, model, singer. If you don't know who Justin Timberlake is, listen to Until the End of Time by Justin Timberlake and thank me later.

HISTORY OF LORA

I made this lora because there's no adequate Justin Timberlake lora models. So, I had to go out and create my own. That's pretty much the gist of it.

I oftentimes use well known people or celebrities as muses for character creation. J.T is happens to be the muse for one of my more recent characters. So I put the time investment into training this model. Wasn't much of a time investment in the end. The training pretty much got it in the first try.

If I rated the likeness rating, I'd say its about 90% in terms of accuracy.

TRAINING INFO

I will add the training info in the version information about this model.

USE CASE

I don't think I really need to say this, but I need to say it. Please, for the love of all that is good, don't abuse this model. This is a real person, with real feelings. Keep your perverted fantasies in your head. My lora isn't here to make it a partial reality.

Be creative and imaginative, but I strictly forbid NFSW content being created with this lora. Obviously you can't post it on Civitai, but have the decency to keep it out of your private collection to. Any lora I create is for the purpose of creativity, not perving out on the celeb of your dreams.

HOW TO USE

I've only done preliminary testing on the lora. Once I have a more fleshed out testing period, I'll add some suggestions on how to use. But I can give my initial testing info.

DPM++ 3m SDE Karras (I use this and 2M Karras, but prefer to use 3M if it doesn't freak out the renders)

40 Steps

Hi-Res fix after testing if the Lora is accurate or broken.

Adetailer: Usually helpful though I've noticed it can skew with results in likeness accuracy. Depends as it's very random when it does.

512 Width
768 Height
This is because the training dataset isn't square but rectangular. So maybe avoid 512x512 or 768x768. Don't know what will happen.

描述:

This is the first version of the Justin Timberlake model. I'm doubtful I could possibly get it more accurate. But there's always room for improvement I guess. I'm certain I might tweak one more model out of the training dataset.
47 Images used for training
AdamW for the Optimizer (Only because I'm using an AMD GPU)
Cosine for the Scheduler
4700 Max Steps
1 Epoch
None: Cross Attention (AMD GPU doesn't allow Xformers use)
BATCH SIZE: 1

512x768 Resolution
NETWORK DIM: 128
NETWORK ALPHA: 96

Maybe this can help you train your own Lora? Maybe there's something here comprehensible since Lora training seems to be very much luck based in my own personal experience.

训练词语: justinTimeberlake