Based on the input image, it generates an image of the SD3.5 Large model, which is automatically tagged by WD14tagger, so it works without prompting. Try varying the value of denoising strength and confidence threshold.