Lora is "slightly" overtrained on some concepts that included only one image, which affects the quality and flexibility. Some concepts only work in conjunction with certain poses. Sometimes it is difficult to change emotions. In this case, lower the strength, or just use Inpainting. It's just a hastily done test with lots of mistakes, don't expect high quality and guaranteed results from it.
Description
trained on 29 images from Pinterest upscaled to 1024X1024, mixed quality. Some pictures turned out to be superfluous and too complex...















