Post #2623587

If the upscaler was trained on data that contains that exact patch in other contexts, it'll theoretically be capable of upscaling it properly.