Type | |
Stats | 265 0 |
Reviews | (21) |
Published | Apr 26, 2024 |
Base Model | |
Training | Steps: 3,000 Epochs: 1 |
Usage Tips | Clip Skip: 2 Strength: 1 |
Trigger Words | amandamanopo |
Hash | AutoV2 9248A88954 |
Description
Hi folks! this LORA is to generate preferably realistic images based on a real person, an Indonesian celebrity, Amanda Manopo. This is my first time training any LORA ever, so do expect updates later on! Generation performance and sample prompts are available below.
Training Parameters
I will only list some things, such as:
Dataset Images: 40
Regularization images: 400
Steps per image: 30
Batch size: 1
Learning Rate: cosine_with_restarts | 3e-6
Text Encoder Rate: 3e-1
Disclaimer
I do not own the dataset images, hence why it is prohibited to sell generated images for any commercial and/or NSFW purposes. I urge everyone to exercise utmost caution and adhere to ethical guidelines, as the individuals depicted are real people.
=============================================
Amanda_Manopo_v1
As the first version of this LORA, I have tested it with several models, and below is the result/performance:
Prompt: photo of a beautiful woman, <lora:Amanda_Manopo_v1:1>
Negative Prompt: ugly, blurry, disfigured
=========
Realisian_v60
perfectdeliberate_v5
cyberrealistic_v42
Currently the best checkpoint to use with the LORA