FOR SDXL VERSION
Trained on the epiCRealism XL model and should therefore be used with this model for perfect use.
I'm looking for photographic rendering only, I'm not interested in 3D or 2D rendering. So my Lora will probably never be trained on base model versions.
I highly recommend using After Detailer.
Trigger token is ohwxtd
Can be used at an intensity of 1.0 but I always prefer to stay at 0.9
Info about train settings :
LoRA trained with a dataset of 107 images / 10 repeats / 10 epoch with batch size at 1.
Mixed and save precision = bf16
Optimizer = AdamW
Resolution = 1024x1024px
FOR SD 1.5 VERSION
Trained on the epiCPhotoGasm-Z model and should therefore be used with this model for perfect use.
But work also very well with RealCartoon-Realistic
I'm looking for photographic rendering only, I'm not interested in 3D or 2D rendering. So my Lora will probably never be trained on base model versions.
I highly recommend using After Detailer.
Trigger token is ohwxtd
Can be used at an intensity of 1.0 but I always prefer to stay at 0.9
Info about train settings :
LoRA trained with a dataset of 65 images / 8 repeats / 13 epoch with batch size at 4.
Mixed and save precision = bf16
Optimizer = AdamW
Resolution = 512x512px
Description
I completely redid the captioning as well as the training's hyperparameters and fine-tuning, which gave me a better overall understanding of Lora training (Thanks to ChatGPT version o3-mini).
So here's a more accomplished and flexible version, I won't be doing any other Thais's versions with SDXL.
The next one may be for Flux
Cheers.