Since somebody asked, and since it'd be a bit of a long answer, here's my thoughts in journal form. I've picked this up from a random assortment of some online guides and experience, so I'm not going to say this is The Right Way or anything like that - just what seems to work for me, at least for now. That caveat out of the way, I have a few different fundamentals I apply: 0) Not really a tip, but make sure your canvas size works right! Try to keep it a multiple of 64 if you're using the model setup I am (i.e. following PseudoChmod's setup guide). If not you'll notice that every round of inpainting actually stretches or compresses the image a tiny bit, eventually getting to a pretty significant degree. (The models are trained on 512, 768, and 1024 images so those are the ideal as I understand it, but I've found going outside that strict triad isn't too big of an issue as long as you keep to that multiple of 64 rule.) 1) Is this one worth inpainting? Sometimes it's a promising image,