It Works with all SDXL models.This is a LoRA model for mosaic correction. To enhance versatility, training was conducted using Multi-ADDiFT with five pairs of images. Since it does not affect anything other than the mosaic in the image, it can be left applied without concern. A recommended strength setting is 1.0.When the intensity is reduced to minus, it will perform de-censoring image processing. It is thought that about -0.5 is just right. If you do more than -1.0, it will generate more detailed genitals than the original image.When using it simultaneously with other LoRAs, please set the intensity to approximately +2.0.<hr />モザイク修正を行うLoRAです。汎用性を高めるため、Multi-ADDiFTを用いて、5ペアの画像を用いてトレーニングを行いました。画像にモザイク以外の影響を与えないため、入れたまま放置が可能です。強度は1.0をおすすめします。強度をマイナスにすると、画像の無修正化を行います。-0.5程度がちょうどいいかと思われます。-1.0以上にすると、元の画像より詳細な性器が生成されます。他のLoRAと同時に使用する場合、強度を+2.0程度にしてください。トレーニングの詳細/Training details (TrainTrain):Mode:Multi-ADDifTimage size(height, width):1024train iterations:120train learning rate:1e-4train batch size:1train optimizer:CAMEnetwork blocks(BASE = TextEncoder): OUT00,OUT01,OUT02image mirroring:Enabled

Files with SHA256: 3f4be9067923a6021fadfb1b01a8ba18bdc7905f062bfe276a4e2897d36541fe