AU (delete_au) Style LoRA

Trained on AnyLoRA, works great on it and on most AOM/Anything derived anime models. Use au (d elete) for more accuracy and weight should be around 1. For the upscaler the effect it gets using Latent at 0.6 denoise.