Some help for accurate prompting

[CharlesOkwuagwu]

Please help me with a more accurate or simpler prompt for this action:

change the clothes of the lady in image_1 to the clothes in image_2

[able2608]

Per my current experience, changing clothes using reference images (given a clothes image and a subject image) does not work that well. Changing the color of the clothes or changing the clothes with text prompt should work, but the parameters (especially both of the guidance scale) need to be tuned and tailored for individual prompts and scenarios. Before we figure out a way to reliably do so, perhaps you would like to use text prompts to describe the clothes of your choice instead.
BTW, if you are editing an image (like changing clothes, removing or adding objects etc.), you might want to put the image tag (<|image_n|>) at the front of the prompt, before you describe the task. This is what the author suggests. Also, avoid using the term "remove" if you are trying to replace an object if something is suppose to be occluded by the object you want to replace, like changing pants to skirts. Using the term "remove" will cause the model to erase the legs altogether, resulting in floating skirt under the torso of the subject.