This is the test version for 80 images. I trained this LoRA using the script modified by AI-toolkit.

The training set for this model is based on the same dataset as "Xiaohong Dou," with some additional data added and trained using FLUX. The concept of the model is to train on clothing and human figures. However, when the training set exceeds a certain number of images, it can lead to composition anomalies. As a result, this model might occasionally generate distorted hands and feet, but most of the time, they are good.

Sampling Method (Sampler): Euler simple

Sampling Iteration Steps: 20

Width: 768, Height: 1280

The modified GUI local trainer is available on my planet 🌎 https://t.zsxq.com/8fA0M

discussion group: QQ 852450267