This is my first try at training a lora from photos. I tried to get good photos of female genitalia.
Description
FAQ
Comments (5)
I have no idea. This was one of my first tries with photographic reference.
@webegm171 Trained on 129 images. about 15 tags on each. I hear flux trains better on simpler captions, but I haven't had another success yet.
If the parameter can be adjusted using Kohya ss to increase the size, it may result in a qualitative improvement, as shown in the following summary
Network dim: Linear dim, representing the size of the model. The larger the value, the finer the model. It is commonly used between 4 and 128. If set to 128, the LoRA model size is 144M.
Network alpha: Linear alpha, usually set to be smaller or the same as Network Dim, usually set Network Dim to 128 and Network alpha to 64.