Type | |
Stats | 50 |
Reviews | (1) |
Published | Feb 3, 2024 |
Base Model | |
Training | Steps: 3,240 Epochs: 30 |
Usage Tips | Clip Skip: 2 Strength: 0.7 |
Trigger Words | Lyoto lyoto "the dragon" machida machida |
Hash | AutoV2 63C5130EAB |
This was mostly a test to create a lora for a specific person and be able to get him into diffrent situations without being overfitted with his original data. I am somewhat pleased but please test yourself, i have tried deciding what color of clothes he should wear, what type of clothes and even merged him with other persons/genders.
I sugest around 0.7 in lora strength and he does best in bigger height then width in my tests. Trained with 30 epocs on 40 images in 2 batches so abit over 3000 iterations.
It is a bit overfitted if i say so but gives great detail to his look, so a couple of tips to be able to use this is to use negative parameters as gloves, logos, sponsors and naked.
Some postive that the data is trained with is fullbody, upperbody, fightpose