Hm, so with my groundbreaking discovery of captions I've tried to rebake the Regloss loras, but surprisingly very little have changed.
It still just refuses to learn some stuff. Like Kanade is supposed to have piano keys on the jacket. It is there on all the training images that include lower body, even added some close-ups but they just refuse to show up. I think I'm doing something wrong, when it can learn weird outfits like this
>>75578685 relatively accurately.
