overtrained model, 8000 steps, 10 epochs, vajar el lora to be more creative. Use ADetailer to improve face, justify that she is a woman with a Japanese face for best results.
Description
The face was reconstructed, I have reached 8 thousand steps, the dim 128, alpha 8. I need to understand what alpha does to improve training. The version has gone up to version 4 because many things have been modified. Adetailer is not needed but the results improve with this. It is important to have a good promise. Anyone who knows how to improve the assertiveness of the face please comment on how the lora has improved. I think there is a better format than LoRa but I don't know how to use it. that GFPGAN, and a rescaler for better results