This slider LoRA changes the clothing from casual style to avant garde style.
As z-image associates the avant-garde style to other parameters such as physical appearance and lighting, these parameters are affected by the LoRA too.
The sample images weights are -3.0 on the right side (Casual), 0.0 in the middle, and 4.0 on the left side (High-End Fashion).
Hope you like it!
Description
FAQ
Comments (2)
Is it possible to correct the problem of applying minus to make the face look Asian and applying plus to make the face look Western?
z-image was clearly trained on "high fashion" images that had a lot of other similar features, like "caucasian person", "hard shadows" or "woman", which makes the training process for this concept LoRA very hard to steer in the right direction.
That said, in all the samples generated for testing I had no issues keeping the ethnicity consistent by just specifying it in the prompt :)