Maria Andrejczyk by Brassen250版本v1.0 (ID: 272878)

About the model? What is there to really tell about it? It's model of Maria Andrejczyk, an Olympian Athlete. I wasn't really familiar with her at first, but now I know a little bit about her and I think she's pretty incredible.

I want to thank the request to create this model. It had me step outside of my box of knowledge and learn something new about someone else. It also forced me to extremely careful about my dataset since finding the right images from the public domain (internet) was actually difficult to do.

Anywho, I think the model came out quite nicely.

HOW TO USE

Um, you know, prompt and stuff. I use 0.7 as a base for the lora strength of any lora and increase or decrease accordingly. Same with this lora. Unfortunately, since the dataset is limited, there are some burned in things despite my best effort to not have to much focus on certain features of the images.

You have to prompt well to avoid ending up in sportswear. It will bleed into your dresses and other outfits.

I've also baked muscular into the prompting, but on purpose. About 25% to 30% of the dataset of images had muscles on display. So... yeah. Negative prompting gonna need to work overtime in case there are some natural features you don't want.

Other than those issues, happy prompting.

USE CASE

Come on, what did you expect to see here? The same rule applies as with all my Loras. Respect... the... person... Don't use this lora to create some private library of perversion. NSFW content with this lora is strictly prohibited. This goes beyond the Civitai posting rules. This includes posting explicit images of this lora on other sites or your own private collection images. Treat others as you want to be treated. This is a real person with real feelings.

描述:

This is information on how the Lora was trained.

DATASET: 54 Images
SCHEDULER: Cosine
OPTIMIZER: AdamW (AMD GPU)
TRAINING RESOLUTION: 512x768 = 2:3 Ratio
NETWORK DIM: 128
NETWORK ALPHA: 64

训练词语: maria andrejczyk