The LoRA did not learn any faces.

The basic idea of this LoRA was to create topless images with certain "characters" without altering the face. Anyway in my experience FLUX LoRAs do this sometimes even if there is zero face data in it.

If you don't have a potato pc i highly recommend using ComfyUI and go way above the classic 812x1236 resolution. There are enough examples in the gallery to drag n' drop.

In general, if your "character" LoRA already has breast size data, this LoRA can adapt. It has learned 1000 different breasts and pretty much every breast size.

It does not work good with the generator on civitai.

It works better on tensor.art also enough examples over there.

Weaknesses:

Lighting: It learned from 1000 images in a studio.

Breasts from the side: It didn't learn many nipples from the side. That is something i intend to correct in a future v3.

v2

Trained for another 8 epochs on the same 1000 images

___________________________________________________________

v1

Trained on 1000 images from 🇨🇿casting for 12 epochs.