This lora was based on the model anythingv4.5 and trained by picture of the sajou manaka in Fate/Prototype, there are many problems in the first attempt, which may be improve after being skilled. It is recommended that the weight be around 0.8 with tag:sajoumanaka,1 girl at the front of the prompt may be better at least that's what I tested.
2023.4.25
Update the new version sajoumanakaV2.0
2023.5.22
Update the new version sajoumanakaV3.0
Description
This time, I tried using the SD1.5 model for training, and the stability of the produced images and characters was significantly improved, but some details were not well represented. Even if images with rei ju are added to the training set, they cannot be restored. Looking for a solution.