Experimental test LoRA for creating anime style images. Appears to be mono-style for the most part.
Works better with detailed prompts.
Recommended LoRA Strength: 0.8-1.2
*Has a chance to create more realistic images due to training issues.
Trained on a rather crude dataset consisting of 55 promptless images from WAI-illustrious-SDXL.
Dataset was captioned by Florence2.
Description
Test LoRA. Produces hallucinations a lot.
I don't recommend going below 0.8 or above 1.2 LoRA strength. From my testing, <0.8 would make more realistic images and >1.2 would increase hallucinations. (0.8-0.9 occasionally makes realistic images)
I have not tested combining it with other LoRA's.
Later versions will be more refined throughout all steps of the training process.