>>3034122detection is just for how sensitive the yolol is, which is irrelevant for inpainting, for everything that affects the actual img2img they're the same process (other than your ability to draw a mask compared to a detection algorithm drawing rectangles over things), other than that i mean you should be prompting your inpaints just like your adetailer, you should be inpainting at equal or higher resolution to the image being inpainted (but again, not so high that it takes 40Gb of ram, 6-8mil pixels is starting to push it for a 3090 without resorting to tiled decode), your denoise, mask blur & pixel padding should be set just like with adetailer
there's nothing particularly advanced going on in the backend with adetailer, it's just an automation layer, nothing wrong with it but there's no reason for it to take less memory for the same job than inpainting other than your settings
