View from directly below [Haruka v2/DiT v1]

The AI can struggle with a view completely from below, often leading to undesirable results. This LoRA aims to fix that, with training images below various types of skirt with various types of underwear. There are also a few of the angle but with shorts or a swimsuit but I think the rest of the dataset will make this best suited to skirts. Trained for Haruka v2. Appears to work fine with hoshino too.