Model Card for LoRA
Based on a 48 image dataset scraped from Danbooru and tagged with WD1.4 Tagger. Trained for 30 epochs (7200 steps), best with models and merges based on Anything v3.
Not particularly prone to NSFW, as the training dataset was somewhat balanced, but is capable of it.
Outfits tend to Ruby's default colors of red and black unless specified otherwise, especially all kinds of dresses.
I also recommend using Latent upscaler with medium (0.4-0.5) denoise, as it can fix some small inconsistencies like wrong eye color.
Model Description
Version 1.0 is currently the only one available and is the least prone to straying from the prompt (white dress stays white), however, may be slightly inaccurate when depicting Ruby.
Best weights seem to be in the area of 0.6 to 0.7, and for best results I recommend adding in tags like "grey eyes, red hair, multicolored hair".
Higher weights can sometimes lead to facial artifacts and/or weird anatomy.
- Developed by: DarkSolus
- Model type: LoRA
- Finetuned from model [optional]: Anything v3
How to Get Started with the Model
Download the preferred version of the LoRA from the repo.
Install Additional Networks extension:
- via Auto1111's extension manager
- via GitHub: https://github.com/kohya-ss/sd-webui-additional-networks
Reload the UI, and place your downloaded LoRA into: .\stable-diffusion-webui\extensions\sd-webui-additional-networks\models\lora