The training set only includes Asian female bodies.

Automatic annotation model used: llama-joycaption-beta-one-hf-llava

Training Method: 1. Destroy nude images to train the model on human anatomy (as same as decensored task). 2. Use illustrated images to train on clothing removal.

For realistic character tasks, the recommended value for flux guidance is 1.0. For anime-related tasks, you can set the value higher.