Suggestions: input person mask?

[wandrzej]

In the original gradio demo, one of the inputs was also person mask, either automated, or provided by the user. In the provided nodes, we only have parameters for image with person, garment mask and garment image. I think it might help quite a bit, as most results get the mask wrong

Another thing that it seems to be performing differently from the original implementation is overall quality. Haven't done any head to head comparisons, but the end effect is like a very quick process, lacking sharpness.