v1.1 Improvements to color and details, weight recommended 0.7 to 1, recommended VAE are Counterfeit and OrangeMix or any NAI renamed.

As usual tried to recreate Marushin's with a VERY SLOW cook with the intent of allowing expressions already present in the checkpoints without too much bleeding from the ones in the dataset, I think this LoRA manages that decently well allowing for a bit more of creative freedom without external LoRAs or LECOs, while also taking other LoRAs and LECOs without much issue, weights go from 0.65 to 1, all pics were made at 1.

Trained for 256 epochs with 138 images at 768x768 resolution, for a total of 49032 global steps (What a long bake, my god)

All preview images have Hires and Adetailer, LoRA likes VectorScope for saturation, and contrast.

Any and all reviews are very much appreciated so I can know what to improve for this LoRA or for other models, thank you~

With this LoRA I've exhausted all the datasets I had ready for cooking, next release might take a while due to collecting images and tagging them, because one lora per day, per style isn't sustainable tehee~